Code of Life
Genomics, proteomics, systems biology โ decoding the blueprint of life.
Nature Genetics๐ค AI summarySAIGE-QTL Overcomes Limitations in Discovering Single-Cell Transcriptomic Variants Using Poisson Mixed Models
Background Expression Quantitative Trait Loci (eQTL) analysis, which identifies the impact of genomic variants on gene expression, has become a standard approach for interpreting the functions of disease-associated variants. Recently, advancements in single-cell transcriptome analysis (scRNA-seq) technology have fueled research into cell-type-specific eQTLs that act only on specific cell subtypes. This is because traditional bulk transcriptome analysis, which involves homogenizing entire tissues, struggles to capture the expression regulatory signals of rare immune cells or specific cell populations within the microenvironment. However, significant statistical barriers existed for accurately calculating the correlation between genetic variants and gene expression at the single-cell level. Read depth varies across cells, and the proportion of zero counts (where expression is not detected) is extremely high. Researchers have primarily adopted a 'pseudobulk' approach, summing expression values of cells per individual to apply linear regression models. This approach significantly dilutes the heterogeneity and intracellular expression variation of individual cells, making it difficult to precisely correct for inter-individual relatedness or the genetic structure of multi-ethnic cohorts. Due to computational complexity, there were also clear limitations in testing the combined effects of low-frequency rare variants. Key Findings Researchers in the United States have reported SAIGE-QTL, a statistical genomics framework that directly handles single-cell read counts without post-processing, in the international journal Nature Genetics. This methodology is designed based on a Poisson Mixed Model to reflect the count data distribution unique to single-cell data without distortion. By modeling the correlations among thousands of cells derived from individual donors and the genetic relatedness between samples as random effects, Type I errors (false positives) are suppressed. This structure directly controls covariates within the model, such as the total transcript count per cell or batch effects. Compared to the pseudobulk approach, statistical power is significantly improved for rare subtypes with low cell counts or genes with low expression levels. The research team achieved computational efficiency capable of handling genomic data from tens of thousands of individuals and millions of single-cell data points by applying approximation estimation techniques validated in large-scale biobank-linked analyses. The most prominent achievement is the implementation of rare variant set-based association testing in a single-cell environment. Because individual rare variants are low in frequency, it is nearly impossible to prove significance at the single-variant level. SAIGE-QTL directly linked burden tests and variance component testsโwhich group rare variants by gene or regulatory region unitsโto single-cell expression levels. Significance and Outlook SAIGE-QTL opens the path for genomic discovery spanning from common variants to rare variants at single-cell resolution. While previous studies were limited to analyzing common variants among single-nucleotide polymorphisms (SNPs), it is now possible to discern the effects of protein loss-of-function variants or rare non-coding variants in promoter regions on specific cell types. It is expected to become a foundational tool for functional genomics research aimed at tracking the cellular pathways in which disease-susceptibility genes cease to function. It is particularly likely to be widely used in the analysis of complex diseases with highly restricted pathogenic cell types, such as neurodegenerative and autoimmune disorders. However, the requirement for massive memory and high-performance computing resources remains a challenge. To comprehensively model spatial information between cells and the physical signals of neighboring cells in spatial transcriptomics data, further algorithmic expansion is required.
๐ก In drug development, the key to reducing clinical failure rates when validating disease targets is human genetic evidence. SAIGE-QTL provides an accurate causal map of how specific genetic variants disrupt the expression of particular immune cell subtypes in patient-derived single-cell data. For example, it can confirm whether an unidentified non-coding rare variant found in a cohort of patients with lupus or rheumatoid arthritis selectively impairs the suppressive function of regulatory T cells. Early elucidation of the cell-specific mechanisms of disease-associated targets can reduce off-target side effects during the candidate discovery phase and enable the precise design of biomarkers to select responder patient populations.
Nature Genetics๐ค AI summaryIntegrating Spatial Multi-omics Data of Different Resolutions into a Single Map Using Graph Attention Deep Learning
Background Spatial biology, which identifies both the location and molecular characteristics of cells within tissues, has advanced rapidly in recent years. Key pillars include spatial transcriptomics, which tracks the location of gene expression at the single-cell level, and spatial proteomics, which captures protein distribution across tissues. The problem lies in the fact that each measurement device has entirely different spatial resolutions and target molecules. Some platforms observe individual genes at the sub-micrometer scale, while others measure thousands of genes in bulk at the spot scale of tens of micrometers. Protein and metabolite analysis instruments also face varying physical resolution limits. Until now, researchers have primarily analyzed heterogeneous data from the same or adjacent tissue sections individually and then manually compared them or overlaid them using simple statistical interpolation. This approach fails to bridge the physical gap between different resolutions and makes it difficult to precisely correct for tissue deformation and batch effects that inevitably occur during measurement. There was a desperate need for a computational algorithm capable of constructing multidimensional spatial maps by registering high-resolution single-cell information with low-resolution molecular maps. Key Findings Researchers at Stanford University have overcome this limitation by constructing CORAL (Cross-scale spatial Omics Registration and Analysis via graph Learning), a deep generative model based on graph deep learning. CORAL is an AI model designed to register and integrate spatial omics data with different resolutions and modalities into a single latent space. It adopts a graph-attention aggregation mechanism as its core computational structure. Researchers transformed the tissue microenvironment into a graph structure, defining cells or measurement spots as nodes and physical spatial distances as edges. The Graph Attention Network (GAT) flexibly collects neighbor information by calculating weights based on the relative distance to surrounding cells and molecular expression similarity. It also demonstrated high precision in the cell deconvolution process, which resolves low-resolution spot signals containing mixed cells into individual cell units by comparing them with high-resolution single-cell data. Experimental results showed that CORAL achieved stable alignment of multi-platform data without physical distortion, even under conditions with severe morphological deformation between sections. Compared to existing benchmark models, both spatial domain clustering consistency and the accuracy of predicting unmeasured molecular expression were found to be excellent. Significance and Outlook The announcement of CORAL provides a computational foundation for unifying fragmented spatial biology experimental data into a single analytical framework. Even if transcriptome and proteome data are collected via separate platforms, they can be reconstructed into high-resolution multi-omics maps through post-algorithmic processing. Researchers have secured a pathway to explore the three-dimensional microenvironment of diseased tissues from various angles by reusing existing accumulated spatial data without incurring massive additional experimental costs. However, practical challenges remain before full-scale implementation in clinical settings and large-scale research. Due to the nature of graph attention computation, whole-slide datasets with hundreds of thousands of cells incur significant memory consumption and computational overhead. Establishing standardized experimental protocols to fully control technical noise and signal deviations across data collection platforms is also a task for follow-up research.
๐ก CORAL can demonstrate immediate utility in analyzing the tumor microenvironment of cancer patients and in drug development. By combining high-resolution transcriptome maps and immune protein distribution data obtained from patient tissue biopsies using CORAL, it becomes possible to precisely track local signal exchanges between cancer cells and infiltrating immune cells at the single-cell spatial level. Even in pathology laboratories lacking expensive, state-of-the-art multiplex staining equipment, the spatial location of target proteins can be accurately estimated by registering standard single-omics data with reference maps. It is also expected to contribute to the three-dimensional elucidation of how drug candidates reach target cell populations within tissues and what responses they induce at the genomic and proteomic levels during the non-clinical stage.
Nature Genetics๐ค AI summaryInformation Theory Framework TCRi, Linking T-cell Genotype and Cell State, Opens New Horizons for Predicting Immunotherapy Response
Background T-cells, the key commanders of the immune system, recognize antigens and form unique cell states through T-cell receptor (TCR) sequences present on their surface. Quantifying the correlation between the genotype (the receptor sequence information of individual T-cells) and the phenotype (represented by gene expression patterns) has been a long-standing challenge. Despite significant advancements in single-cell transcriptome analysis and TCR repertoire sequencing technologies, it has been difficult to link the random diversity of sequences and complex cell differentiation pathways with a single computational formula. Existing studies have been limited to methods such as counting the proliferation frequency of specific clones or simply comparing the expression levels of known differentiation markers. These static analyses fail to explain the fundamental causal role of receptor sequences in determining cell state diversity. Because the reasons for the drastic variation in T-cell responses among patients undergoing anticancer immunotherapy could not be elucidated at the cellular sequence level, the discovery of effective biomarkers has also reached its limits. Key Findings Researchers have constructed TCRi, a computational framework that integrates the concept of mutual information (MI) from information theory into immunogenomics. TCRi precisely calculates the shared non-linear information between the receptor sequence variations of individual T-cells and their transcriptome profiles. It is an algorithm that directly measures, in a non-parametric manner, how much the genotypic sequence reduces the uncertainty of the cellular phenotype. Analyzing integrated datasets of single-cell RNA sequencing (scRNA-seq) and single-cell TCR sequencing (scTCR-seq) with TCRi revealed that the mutual information between T-cell receptor sequences and cell states is a key indicator determining the response to immune checkpoint inhibitor therapy. Patient groups exhibiting high mutual information showed significant correlations with long-term survival rates and tumor shrinkage rates following immunotherapy. Compared to existing indicators such as clonal expansion or simple cell differentiation indices, the mutual information index calculated by TCRi demonstrated clear discriminative power in predicting patient treatment prognosis. Significance and Outlook This study represents a breakthrough in resolving the genotype-phenotype network of immune cellsโa longstanding challenge in complex systems biologyโby reducing it to mathematical information theory. By quantifying the deterministic impact of T-cell sequences themselves on cell fate, it has established a computational framework applicable not only to immuno-oncology but also to research in autoimmune diseases and infectious diseases. It lays the foundation for advancing algorithms capable of gauging the immune response potential of a cell using only the receptor sequence. For actual clinical entry, further validation work in various cancer types and large-scale patient cohorts must follow. Challenges remain in correcting errors in mutual information calculation due to sequencing depth and sample collection timing, and in establishing a standardized pipeline that can be immediately integrated into clinical practice.
๐ก The TCRi framework serves as a next-generation precision diagnostic algorithm for the early screening of responsiveness to anticancer immunotherapy. By analyzing T cells in a patientโs blood or tumor tissue at the single-cell level to derive mutual information values prior to prescribing immune checkpoint inhibitors, it enables the pre-identification of drug-refractory patients, thereby reducing unnecessary treatment costs and immune-related adverse events. It can also be utilized as a critical quality indicator in the development process of cell therapies. When manufacturing patient-specific CAR-T or TCR-T therapies, it is possible to design a pipeline that dramatically enhances the in vivo persistence and anticancer activity of the therapeutic agent by selecting the optimal T-cell clone state capable of maximizing therapeutic efficacy using mutual information metrics.
Nature Genetics๐ค AI summaryReducing Intensity of Pediatric Brain Tumor Treatment with Multimodal AIโฆ Protecting Survival Rates while Minimizing Lifelong Sequelae
Background Pediatric brain tumors are fatal diseases, ranking as the leading cause of death related to pediatric cancer. Children with central nervous system tumors, including Medulloblastoma and Ependymoma, undergo standard treatments involving high-dose radiation therapy and intensive cytotoxic chemotherapy following surgery. Thanks to these aggressive treatment strategies, the 5-year survival rate has significantly improved compared to the past. The problem is that whole-brain and spinal radiation, along with high-dose chemotherapy administered during childhoodโwhen the brain is not yet fully developedโcan leave irreversible permanent disabilities. Many pediatric patients who complete treatment suffer from severe cognitive decline, learning disabilities, and growth hormone deficiency, and face an increased risk of secondary cancers as adults. In clinical settings, there has been an urgent need for therapy de-escalation strategies that can safely reduce treatment intensity without lowering survival rates. Due to the vast variation in prognosis among patients and high molecular genetic heterogeneity of tumors, it has been difficult to accurately identify low-risk patients suitable for de-escalated treatment using only existing histological staging or single-gene testing. Key Findings Researchers have constructed a multimodal AI-based prognostic prediction model capable of integrated analysis of vast clinical records and molecular profiles. The core of this technology is calculating personalized, time-dependent survival risks rather than relying on single information points at the time of diagnosis. Integration of Multimodal Data and Time-Dependent Risk Estimation The model was trained by combining imaging data such as Magnetic Resonance Imaging (MRI), DNA methylation profiling, transcriptomic gene expression data, and basic clinical pathology information. Unlike existing scoring systems that were limited to static risk stratification, this model reconstructs survival probability curves in real time for specific time points. The multimodal algorithm precisely separated extremely aggressive subtypes from relatively benign molecular subtypes even within tumors that appeared histologically similar. As a result, it identified the group of patients eligible for therapy de-escalationโthose capable of long-term survival without excessive high-dose radiationโwith higher accuracy than conventional methods. Identifying the Risk of Overtreatment The study revealed that some patients who would have been classified as high-risk and required whole-brain radiation therapy to the entire brain under existing clinical guidelines actually belong to a molecularly very stable, low-risk group. Conversely, it accurately captured hidden high-risk patients who were judged as low-risk via biopsy but had a high potential risk of recurrence, necessitating intensified early treatment. Significance and Outlook This research marks a new turning point in precision oncology, considering the quality of life of pediatric cancer survivors after treatment. It moves beyond the stage of simply removing cancer cells and saving children, as it uses AI to provide clinical evidence for minimizing the neurological and physical sequelae that a child may experience throughout their life into adulthood. To accelerate clinical application, several practical challenges must be addressed. Since pediatric brain tumors are rare diseases, there are limitations to using data from a single institution; therefore, external prospective validation through multinational and multi-center cohorts must precede implementation. A process of refining the model's generalization performance is also required to overcome variations in MRI protocols and genomic sequencing platforms between institutions. Nevertheless, this architecture, which integrates data from various biological layers to predict prognosis over time, holds great potential for expansion into the treatment design of adult cancers such as gliomas or other pediatric solid tumors.
๐ก This model can serve as an immediate decision-support tool in multidisciplinary pediatric brain tumor clinical settings. When neurosurgeons, pediatricians, and radiation oncologists convene to determine the scope of adjuvant therapy post-surgery, inputting a patientโs multimodal data visualizes the time-specific progression-free survival rate and the risk-benefit ratio of treatment de-escalation. In actual treatment planning stages, for candidates identified by AI as safe for de-escalation, clinical designs such as reducing radiation doses by 20โ30% or localizing the radiation field to the primary tumor site can be confidently applied. This leads to tangible benefits in preventing permanent white matter damage and a decline in intelligence quotient (IQ) in growing children. Conversely, for patients where potential recurrence risk is detected, it becomes possible to design customized protocols, such as early introduction of drug combinations or preemptive targeted therapies, to prevent recurrence.
BMC microbiology๐ค AI summaryBivalent mRNA Vaccine Candidate Targeting Mycoplasma and Chlamydia Pneumonia Designed via Immunoinformatics
Background Community-Acquired Pneumonia (CAP) is a major acute respiratory disease threatening global public health. In clinical settings, infections by atypical pathogens are frequently confirmed, making the identification of causative agents difficult. Representative causative agents, Mycoplasma pneumoniae and Chlamydia pneumoniae, typically begin like a common cold but can escalate into severe pneumonia. Cases of co-infection due to simultaneous exposure to both bacteria are not uncommon, and the resulting complex symptoms have frequently led to difficulties in antibiotic prescription and establishing treatment directions. Currently, there are no approved vaccines globally that simultaneously prevent both pathogens. The previous approach of manufacturing separate single vaccines for each strain has imposed burdens such as causing immune interference between antigens or increasing production costs. As an alternative, the multi-epitope-based mRNA platform is gaining attention. This is because it can evenly activate both humoral and cellular immunity by precisely placing the core antigenic epitopes within a single polypeptide chain. Computer algorithm-based immunoinformatics computation has emerged as a powerful tool for rapidly identifying safe vaccine candidates by screening extensive strain sequences. Key Findings The researchers identified epitopes with over 90% conservation across the analyzed strain sequences. A total of 11 Cytotoxic T-Lymphocyte (CTL) epitopes, 5 Helper T-Lymphocyte (HTL) epitopes, and 14 Linear B Lymphocyte (LBL) epitopes were derived. The selected epitopes were connected with immunostimulatory sequences and linkers to complete two vaccine candidate sequences, MCV1 and MCV2. The two sequences were designed to compare differences in immune activation pathways by arranging different immune-stimulating peptide modules at the N-terminus. Toxicity and allergy evaluations showed that both substances are safe for human cells, and they also exhibited physicochemical stability in terms of the protein's tertiary structure. In molecular docking experiments to verify binding stability with human immune receptors, favorable binding profiles were observed. When applying the precision docking tool HADDOCK, the cluster binding score for the MCV1 and Toll-Like Receptor 2 (TLR2) complex was calculated as -394.3 ยฑ 7.3. The score for the MCV2-TLR4 complex of the exploratory model was also -399.0 ยฑ 5.5, indicating tight binding to the receptor. In Molecular Dynamics (MD) simulations repeated three times for 100 nanoseconds (ns), the conformation remained stable without fluctuation. The binding free energies (MM-PBSA) of the MCV1-TLR2 and MCV2-TLR4 complexes were calculated to be -125.19 ยฑ 20.13 kcal/mol and -126.87 ยฑ 12.73 kcal/mol, respectively. The results of the virtual immune simulation (C-ImmSim) showed active B-cell antibody production and memory T-cell responses occurring simultaneously. The predicted inclusivity index of the global population reached 97.90%. Codon optimization to enhance human cell expression efficiency and the design of a stable mRNA secondary structure were also completed. Significance and Outlook The vaccine design technique applying immunoinformatics has achieved the result of shortening the initial antigen discovery period, which used to take years of manual experimentation, to within a few weeks. Since sequences with over 90% conservation were used as a common reference, it is expected to form a barrier that prevents immune evasion even against pathogen mutations. Arranging the antigen information of Mycoplasma and Chlamydia, which belong to different genera, harmoniously into a single mRNA molecule is regarded as an advanced engineering achievement in the field of multivalent vaccine development. This achievement is inherently limited by its nature as an in silico study based on advanced software prediction and physical simulation. Whether the binding affinity and antibody formation prediction profiles derived from computer screens will function identically in actual living tissues is a separate issue. To ensure efficacy as an actual vaccine, in vivo efficacy evaluations using infected animal models must follow. Securing preclinical data to demonstrate the formulation and in vivo delivery stability of lipid nanoparticles (LNPs) is expected to be a critical milestone for commercialization.
๐ก The bivalent mRNA design proposed in this study provides a concrete solution for clinical and industrial applications in the management of complex respiratory infectious diseases. Because Mycoplasma and Chlamydia have initial symptoms similar to a cold, rapid diagnosis in the field is difficult, and in children or the elderly, co-infection can easily lead to severe complications such as acute respiratory distress syndrome. Pharmaceutical and biotech companies can immediately commence synthetic biology production processes based on the optimized codon sequences and epitope arrangement information derived from the paper. If a vaccine that simultaneously protects against two bacterial pathogens with a single dose is commercialized, it is expected to lower ICU admission rates and significantly reduce the financial burden on the healthcare system.
STAR protocols๐ค AI summaryFrom Nucleases to Prime Editing: CleanFinder Protocol for Precise Analysis of Heterogeneous Gene Editing Results
Background Since the development of CRISPR-Cas9 nucleases, the scope of applications has rapidly expanded to include Base Editing (BE) for precise single-nucleotide correction, Prime Editing (PE) for direct DNA writing and erasing, and mitochondrial genome editing technologies. While attempts to target specific sequences continue in laboratories worldwide, actual cellular responses are far from simple. Cell populations following gene editing often form complex allele mixtures, including random insertions/deletions (indels), unintended adjacent base substitutions, and partial inversions, rather than the intended single outcome. Amplicon sequencing technology is used as the standard analytical method to identify these heterogeneous editing results. However, existing bioinformatics tools are often biased toward specific editing methods or rely on Linux command-line interfaces (CLI), creating high entry barriers for general biologists. Furthermore, uploading large-scale FASTQ sequencing data to overseas cloud servers frequently raises concerns regarding the leakage of patient genetic information and network bottlenecks. There have also been persistent criticisms that these tools fail to objectively filter out allelic dropoutโa phenomenon where specific alleles are lost during PCR amplificationโthereby distorting actual mutation frequencies. Key Findings A research team led by Dr. Alessandro Rossi of the Leibniz Research Institute for Environmental Medicine (IUF) in Germany has published detailed analytical procedures for the genome editing analysis platform 'CleanFinder' in the Cell Press journal STAR Protocols. Alessandro Rossi at the Leibniz Institute for Environmental Medicine (IUF) in Germany published the detailed analytical procedures for 'CleanFinder,' a genome editing analysis platform, in the Cell Press journal 'STAR Protocols.' The team provides systematic guidance covering the entire process, from constructing reference sequence data and FASTQ preprocessing to web browser-based execution and large-scale command-line processing. CleanFinder employs a constrained semi-global alignment method. Based on this, it identifies micro-deletions and point mutations in mitochondrial targets, as well as multiple variants from prime editing, with high precision. The research team has formalized step-by-step interpretation criteria to objectively derive genotype distribution curves, DNA repair patterns, and editing efficiencies. As a key device for enhancing reliability, an allele detection module utilizes heterozygous single nucleotide polymorphisms (SNPs) as markers. This module identifies allelic dropout caused by amplification errors to prevent statistical distortion. The protocol also includes detailed instructions for using a high-speed computation (Turbo) mode to support large-scale screening. The analysis speed and reproducibility of this mode have already been verified in large-scale experiments involving the treatment of 1,849 small molecule compounds on human embryonic kidney cells (HEK293T). Significance and Outlook The greatest advantage of the CleanFinder protocol is the coexistence of data security and analytical accessibility. All computations are completed using only the web browser engine of a personal computer without the need for separate software installation. This structure allows sensitive patient-derived cell sequencing data to be processed on internal laboratory equipment without being exported to external networks. Thanks to its user-friendly interface, laboratories without professional bioinformaticians can obtain high-quality visualization graphs and quantified data within minutes. The protocol is expected to contribute to standardizing the Quality Control (QC) standards required in preclinical and clinical research for Cell and Gene Therapies (CGT). It will also significantly accelerate the selection of candidate substances by allowing various editing platforms to be compared within a single analytical framework. Technical challenges remain. There are clear physical limitations to single-amplicon analysis when attempting to capture large-scale structural variants, such as large chromosomal deletions or translocations of several kilobases (kb) or more. Future expansion into next-generation analytical protocols that organically integrate with long-read sequencing platforms is necessary.
๐ก The CleanFinder protocol provides an immediate quality assurance tool for cell and gene therapy developers and clinical researchers. In the production of Chimeric Antigen Receptor T-cell (CAR-T) therapies, there is an increasing trend of performing simultaneous multiple gene knockouts and therapeutic gene insertions to prevent immune rejection. In such cases, it is impossible to ensure the safety of therapeutic agents if micro-mutations occurring differently at each target site and unintended bystander mutations cannot be accurately tracked. Applying CleanFinder in Good Manufacturing Practice (GMP) facilities allows for the rapid cross-verification of raw data from on-site sequencers without external server connections. By tracking heterozygous SNPs to exclude allelic dropout errors, it becomes easier to strictly control quality bias between batches. This is evaluated to significantly advance the commercialization timeline of new drugs by reducing the costs of generating genome integrity reports for clinical trials and shortening the duration of outsourced analysis.
Nature Genetics๐ค AI summaryUK ๋ฐ์ด์ค๋ฑ ํฌ 49๋ง ๋ช ์ ์ฅ ์ ์ ์ฒด๋ก ๊ตฌ์ถํ ์ฐธ์กฐ ํจ๋, ์ธํจํ ์ด์ ์ ํ๋์ ์์ ์ค๋ฅ์จ ๋ํญ ๊ฐ์
๋ฐฐ๊ฒฝ ์ ์ฅ ์ ์ ์ฒด ์ฐ๊ด์ฑ ๋ถ์(GWAS)์ ๋ณตํฉ ์งํ์ ์ ์ ์ ์์ธ์ ๊ท๋ช ํ๋ ํต์ฌ ๋๊ตฌ๋ก ์๋ฆฌ ์ก์๋ค. ์ฐ๊ตฌ์๋ค์ ๋น์ฉ์ ์ ๊ฐํ๊ธฐ ์ํด ๋ชจ๋ ๋์์์ ์ ์ฅ ์ ์ ์ฒด ์ํ์ฑ(WGS)์ ์ํํ๋ ๋์ , ํน์ ์์น์ ๋จ์ผ์ผ๊ธฐ๋คํ์ฑ(SNP)๋ง์ ์ฝ์ด๋ด๋ ๋ง์ดํฌ๋ก์ด๋ ์ด(microarray) ์นฉ์ ๋๋ฆฌ ์ฌ์ฉํ๋ค. ์นฉ์ผ๋ก ๋ถ์ํ์ง ๋ชปํ ๋๋จธ์ง ์์ฒ๋ง ๊ฐ์ ๋ฏธ์ธก์ ๋ณ์ด๋ ์ด๋ฏธ ์ผ๊ธฐ์์ด์ด ์์ ํ ํด๋ ๋ ๋๊ท๋ชจ ์ฐธ์กฐ๊ตฐ์ ๋ฐํ์ผ๋ก ํต๊ณ์ ์ผ๋ก ์ถ์ ํด ์ฑ์ ๋ฃ๋๋ค. ์ด ๊ณผ์ ์ ์ ์ ์ํ ์ธํจํ ์ด์ (genotype imputation)์ด๋ผ ๋ถ๋ฅธ๋ค. ์ธํจํ ์ด์ ์ ์ ํ๋๋ ์ฐธ์กฐ ํจ๋์ ํ๋ณธ ํฌ๊ธฐ์ ๋ค์์ฑ์ ์ง์ ์ข์ฐ๋๋ค. ๊ธฐ์กด ์ฐ๊ตฌ์์ ๋๋ฆฌ ์ฐ์ด๋ 1000 ์ ์ ์ฒด ํ๋ก์ ํธ(1000 Genomes Project)๋ Haplotype Reference Consortium(HRC) ํจ๋์ ์ฐธ์ฌ์ ์๊ฐ ์์ฒ ๋ช ์์ ์๋ง ๋ช ์์ค์ ๋จธ๋ฌผ๋ ๋ค. ์ด๋ก ์ธํด ํฌ๊ท ๋ณ์ด(rare variants)๋ฅผ ์ ํํ๊ฒ ์์ธกํ๋ ๋ฐ ๋๋ ทํ ํ๊ณ๊ฐ ์กด์ฌํ๋ค. ๋๋ถ์ด ๋ถ๋ชจ๋ก๋ถํฐ ๋ฌผ๋ ค๋ฐ์ ๋๋ฆฝ์ ์ ์์ ์ผ์์ฒด์ ๋ฐฐ์ด ์ํ๋ฅผ ๊ฒฐ์ ํ๋ ์์ ๊ฒฐ์ (phasing) ๊ณผ์ ์์ ์ค๋ฅ๊ฐ ๋์ ๋๋ฉด์, ์งํ ์ฐ๊ด ๋ณ์ด๋ฅผ ๋ฐ๊ตดํ๋ ํด์๋๊ฐ ๋จ์ด์ง๋ ๋ฌธ์ ๊ฐ ์ง์๋๋ค. ํนํ ์ ๋ฝ๊ณ ์ด์ธ์ ๋น์ ๋ฝ๊ณ ์ธ๊ตฌ ์ง๋จ์์๋ ์ฐธ์กฐ ๋ฐ์ดํฐ ๋ถ์กฑ์ผ๋ก ๊ฒฐ์ธก์น ๋ณต์ ์ ํ๋๊ฐ ํฌ๊ฒ ๋ฎ์์ ธ ์ ์ ์ฒด ์ฐ๊ตฌ์ ๋ถํ๋ฑ์ด ์ฌํ๋๊ธฐ๋ ํ๋ค. ํต์ฌ ๋ฐ๊ฒฌ ์ฅ์คํผ๋ ๋ํ๊ต์ ๋ธ๋ก๋ ์ฐ๊ตฌ์๋ฅผ ๋น๋กฏํ ๊ตญ์ ๊ณต๋ ์ฐ๊ตฌ์ง์ ์๊ตญ ๋ฐ์ด์ค๋ฑ ํฌ(UK Biobank) ์ฐธ์ฌ์ 490,319๋ช ์ ๊ณ ์ฌ๋ WGS ๋ฐ์ดํฐ๋ฅผ ๊ฐ๊ณตํด ์ฌ์ ์ต๋ ๊ท๋ชจ์ ๋ฐ์์ฒด ์ฐธ์กฐ ํจ๋์ ์์ฑํ๋ค. ์ฐ๊ตฌ์ง์ 50๋ง ๋ช ์ ๊ฐ๊น์ด ๋ฐฉ๋ํ ์ํ์ฑ ์ฝํธํธ์์ ์์ต ๊ฐ์ ์ ์ ๋ณ์ด๋ฅผ ์ ๋ณํ๊ณ , ์ฒจ๋จ ์ฐ์ฐ ์๊ณ ๋ฆฌ์ฆ์ ์ ์ฉํด ์ผ์์ฒด๋ณ ์์์ ์ ๋ฐํ๊ฒ ์ฌ๊ตฌ์ฑํ๋ค. ์ด๋ฒ ํจ๋์ ๊ธฐ์กด HRC ๋ฐ TOPMed ํจ๋ ๋๋น ์์ ์ ํ ์ค๋ฅ์จ(switch error rate)์ ๋์ ๋๊ฒ ๋ฎ์ท๋ค. ์ธ๊ตฌ ์ง๋จ ๊ฐ ์ ์ ์ ๊ฑฐ๋ฆฌ์ ํฌ๊ท ๋ณ์ด ๋น๋๋ฅผ ์ ๋ฐ ๋ฐ์ํ ๊ฒฐ๊ณผ, ๋๋ฆฝ์ ์ ์ ๋น๋๊ฐ 0.1% ๋ฏธ๋ง์ธ ๊ทนํฌ๊ท ๋ณ์ด ์์ญ์์๋ ์ ์ ์ํ ์ถ์ ์ผ์น๋(rยฒ)๊ฐ ํฐ ํญ์ผ๋ก ์์นํ๋ค. ์ฐ๊ตฌํ์ ์ ๋ฝ๊ณ๋ฟ๋ง ์๋๋ผ ์ํ๋ฆฌ์นด๊ณ, ๋์์์๊ณ, ๋จ์์์๊ณ ๋ฑ ๋ค์ํ ํํต์ ์ง๋ ์ง๋จ์ ์ ์ ํ ๋ง์ดํฌ๋ก์ด๋ ์ด ๋ฐ์ดํฐ์ ์ ํจ๋์ ๋์ ํด ๊ฒ์ฆ์ ๋ง์ณค๋ค. ๊ทธ ๊ฒฐ๊ณผ ๋ชจ๋ ์ธ๊ตฌ ์ง๋จ์์ ๊ธฐ์กด ๋๊ตฌ ๋๋น ์ผ๊ด๋๊ฒ ํฅ์๋ ๋ณ์ด ๋ณต์์จ์ ๊ธฐ๋กํ๋ค. ๋น์ ๋ฝ๊ณ ํ๋ณธ์์ ๊ด์ฐฐ๋๋ ์ธํจํ ์ด์ ์ ๋ณด ์ ์(info score) ์ ํ ํ์ ์ญ์ ๋ํญ ์ํ๋๋ค. ๋๊ท๋ชจ ์ํ์ฑ ๋ฐ์ดํฐ ์ฒ๋ฆฌ ์ ๋ณ๋ชฉ์ผ๋ก ์ง์ ๋๋ ์ฐ์ฐ ๋น์ฉ๋ ์ต์ ํ ํ์ดํ๋ผ์ธ์ผ๋ก ํด๊ฒฐํ๋ค. ์ฐ๊ตฌํ์ ๋ฐ์์ฒด ํ์ ํธ๋ฆฌ๋ฅผ ํจ์จํํด ์์ญ๋ง ๋ช ์ ์ฐธ์กฐ๊ตฐ์ ๊ฒ์ํ ๋ ์์๋๋ ๋ฉ๋ชจ๋ฆฌ ์ ์ ์จ์ ์ค์์ผ๋ฉฐ, ์ ์ฅ ์ ์ ์ฒด ์์ค์ ์ธํจํ ์ด์ ์์ ์๊ฐ์ ๊ธฐ์กด์ ์ ๋ฐ ์ดํ๋ก ๋จ์ถํ๋ค. ์๋ฏธ์ ์ ๋ง ์ด๋ฒ ์ฐ๊ตฌ๋ ์ ์ธ๊ณ ์๊ณผํ๊ณ๊ฐ ์ถ์ ํด ์จ ๊ธฐ์กด ๋ง์ดํฌ๋ก์ด๋ ์ด ๊ธฐ๋ฐ ์ฝํธํธ ์ฐ๊ตฌ์ ๊ฐ์น๋ฅผ ๋จ์จ์ ๋์ด์ฌ๋ ธ๋ค. ์ด๋ฏธ ์ ๋น์ฉ ์นฉ ๋ฐ์ดํฐ๋ง ํ๋ณดํด ๋ ์๋ฐฑ๋ง ๋ช ๊ท๋ชจ์ ๊ธฐ์กด ์์ยท์ญํ ๋ฐ์ดํฐ๋ฒ ์ด์ค์ ์ ํจ๋์ ์ ์ฉํ๋ฉด, ๊ฐ๋น์ผ ์ฌ์ํ์ฑ ์์ด๋ ๊ณ ์ฌ๋ WGS์ ์คํ๋ ์ ๋ฐ๋๋ฅผ ํ๋ณดํ ์ ์๋ค. ํฌ๊ท ๋ณ์ด๊ฐ ์ฌํ๊ด ์งํ์ด๋ ์ ๊ฒฝ๊ณ ํดํ์ฑ ์งํ ๋ฐ๋ณ์ ๋ฏธ์น๋ ๊ธฐ์ ์ ๊ท๋ช ํ๋ ํ์ GWAS ์ฐ๊ตฌ์ ๊ฐ๋ ฅํ ๋๋ ฅ์ ์ ๊ณตํ ๊ฒ์ผ๋ก ๊ธฐ๋๋๋ค. ๋ค๋ง ๋ณธ ํจ๋์ด ์ง๋ ๊ตฌ์กฐ์ ํ๊ณ๋ ๋ช ํํ๋ค. UK Biobank ์ฝํธํธ๋ ์ฌ์ ํ ์๊ตญ ๊ฑฐ์ฃผ์์ ์ ๋ฝ๊ณ ํํต์ ๋น์จ์ด ์๋์ ์ผ๋ก ๋๋ค. ๋น์ ๋ฝ๊ณ ์ธ๊ตฌ ์ง๋จ์ ์ธํจํ ์ด์ ์ฑ๋ฅ์ด ๊ธฐ์กด๋ณด๋ค ๊ฐ์ ๋์ผ๋, ์ํ๋ฆฌ์นด๋ ๋ผํด์๋ฉ๋ฆฌ์นด ์์ฃผ๋ฏผ ๋ฑ ์ ์ ์ ๋ค์์ฑ์ด ๊ทนํ ๋์ ์ง๋จ ๊ณ ์ ์ ํฌ๊ท ๋ณ์ด๋ฅผ ์จ์ ํ ํฌ๊ดํ๊ธฐ์๋ ํ๊ณ๊ฐ ๋ฐ๋ฅธ๋ค. ํฅํ ๊ธ๋ก๋ฒ ์ธ๊ตฌ์ ์ ์ ์ ๋ค์์ฑ์ ๊ท ํ ์๊ฒ ๋ฐ์ํ๋ ์ง์ญ๋ณ ๋ง์ถคํ ๋ฐ์ด์ค๋ฑ ํฌ ๋ฐ์ดํฐ์์ ๊ฒฐํฉ์ด ์ง์์ ์ธ ๊ณผ์ ๋ก ๊ผฝํ๋ค.
๐ก ์ ์ฝ ๊ฐ๋ฐ๊ณผ ์ ๋ฐ ์๋ฃ ๋ถ์ผ์์ ๋ค์ ์ ์ ์ํ ์ ์(PRS)์ ์์ธก ์ ํ๋๋ฅผ ๋์ด๋ ์ค์ง์ ๊ณ๊ธฐ๊ฐ ๋ง๋ จ๋๋ค. ์ ์ฝ์ฌ๋ ์์์ํ ํ์๊ตฐ์ ์ ๋น์ฉ ์นฉ ๋ฐ์ดํฐ๋ก๋ถํฐ ๊ณ ๋ํ๋ ์ฐธ์กฐ ํจ๋์ ์ด์ฉํด ์ฝ๋ฌผ ๋ฐ์์ฑ์ด๋ ๋ถ์์ฉ๊ณผ ์ฐ๊ด๋ ํฌ๊ท ์ ์ ๋ณ์ด๋ฅผ ํจ์ฌ ์ ๋ฐํ๊ฒ ์ญ์ถ์ ํ ์ ์๋ค. ์ด๋ ์์ ์ง์ ์ ํ์ ์ ๋ณ ํจ์จ์ ๋์ด๊ณ ํ๋ณด๋ฌผ์ง์ ์คํจ ํ๋ฅ ์ ๋ฎ์ถ๋ ๋ฐ ๊ธฐ์ฌํ๋ค. ์ง๋จ ํ์ฅ์์๋ ์ ์ ์ฒด ์นฉ ๊ธฐ๋ฐ ๊ฒ์ฌ์ ์ง์ ๋์ฝ์ด ๊ฐ๋ฅํ๋ค. ํฌ๊ท ์ ์ ์งํ ๋ณด์ธ์ ์คํฌ๋ฆฌ๋์ด๋ ๋ณตํฉ ์งํ ๋ฐ๋ณ ์์ธก์์ ์์์ฑ๊ณผ ์์์ฑ์ ์ค์์ผ๋ก์จ, ๊ณ ๋น์ฉ ์ํ์ฑ ๊ฒ์ฌ๋ฅผ ์ ์ ์ํํ๊ธฐ ์ด๋ ค์ด ์๋ฃ ํ๊ฒฝ์ ๊ฒฝ์ ์ ์ด๊ณ ์ ํํ ์ ์ ์ง๋จ ๊ฒฝ๋ก๋ฅผ ์ ์ํ๋ค.
Cancer treatment and research communications๐ค AI summaryDecoding the Crotonylation Landscape of Renal Cell Carcinoma Genes via Machine Learning and Discovery of a Novel Target, CLCNKA
Background Clear cell renal cell carcinoma (ccRCC), which accounts for the majority of kidney cancers, often goes undetected until late-stage metastasis due to its lack of early symptoms. Once tumors progress beyond a certain stage, they frequently become resistant to conventional chemotherapy or radiation therapy, limiting treatment options. Although targeted therapies, including tyrosine kinase inhibitors (TKIs), and immune checkpoint inhibitors have been introduced as standard first-line treatments, the development of drug resistance and interindividual variability in response remain significant challenges that are yet to be fully resolved. This is because patients exhibit diverse tumor microenvironments and genomic alterations that contribute to treatment failure. This is why there is an unceasing clinical need to identify precise prognostic indicators and discover new therapeutic targets. Recently, the molecular biology community has begun to focus on lysine crotonylation (LCR), a type of post-translational modification (PTM). This epigenetic mechanism, in which crotonyl groups bind to lysine residues of both non-histone and histone proteins, is deeply involved in regulating gene transcription and reprogramming cellular metabolism. Although reports have indicated that LCR signals regulate cancer cell proliferation and invasion in various solid tumors, including colorectal and liver cancers, the pathological functions and clinical value of LCR-related molecules in ccRCC remain shrouded in mystery. There has been an urgent need to elucidate the interaction network of LCR genes within vast omics data and link it to therapeutic strategies. Key Findings The researchers systematically profiled multi-omics data for ccRCC by collecting open cancer databases. First, consensus clustering was performed based on the expression patterns of 18 genes directly associated with LCR (LCR-related genes, LCRGs). The analysis revealed that the tumor patient group was divided into two distinct molecular subtypes, C1 and C2. The two subtypes exhibited marked contrasts not only in overall patient survival rates but also in patterns of tumor immune cell infiltration and the activity of major cancer-related signaling pathways. To control the complexity of biological signals, key hub gene modules were extracted using weighted gene co-expression network analysis (WGCNA). Subsequently, 101 machine learning algorithm combinations were validated stepwise to establish a 15-gene-based LCR signature (LCR-related signature, LCRS) model with the highest predictive performance. The completed LCRS model was statistically proven to be a strong prognostic factor functioning independently of existing clinical stages in multiple independent cohorts, including the Cancer Genome Atlas (TCGA). Precision analysis of single-cell RNA sequencing (scRNA-seq) data revealed that the core genes constituting LCRS are predominantly expressed in endothelial cells within the tumor microenvironment. This demonstrates that the LCR regulatory mechanism directly intervenes not only in cancer cells themselves but also in tumor angiogenesis and vascular function maintenance. In drug response evaluation using pharmacogenomic databases (CTRP, PRISM), opposing treatment sensitivities were observed across risk groups. High-risk patients exhibited high sensitivity to sunitinib, whereas low-risk patients showed superior responses upon administration of axitinib and pazopanib. The Cancer Dependency Map (DepMap) large-scale CRISPR screening data identified CLCNKA, which encodes a chloride ion channel protein, as a critical target molecule that suppresses tumor growth in high-risk patients. Additionally, 10 promising candidate compounds targeting CLCNKA for high-risk patients were selected. Significance and Prospects This study comprehensively reconstructed the LCR signaling system, an unexplored area, using computational biology and machine learning techniques to derive clinical value. In particular, the 15-gene model derived from validating over 100 algorithm combinations provides a solid foundation for developing molecular diagnostic panels to stratify risk in patients with small cell lung cancer. By elucidating the expression patterns centered on vascular endothelial cells, the possibility of a multifaceted combination therapy to control the tumor microenvironment has also been opened. Data establishing the differences in sensitivity among TKI agents are expected to significantly reduce trial and error at the prescription site. The finding that drug responses vary depending on a patient's LCR activity, even within the same class of vascular endothelial growth factor receptor (VEGFR) inhibitors, offers highly realistic insights from a precision medicine perspective. The discovered CLCNKA target can also serve as a strong starting point for the development of first-in-class drugs aimed at refractory patient populations who do not respond to existing treatments. However, since the results were derived based on a retrospective multi-omics cohort analysis and public database screening, additional validation is required before practical implementation. The clinical validity of the 15-gene signature must be reconfirmed through prospective clinical trials, and the actual molecular mechanism of the CLCNKA gene within tumors and the in vivo safety of candidate compounds also need to be rigorously evaluated. If subsequent studies smoothly demonstrate the specific pathway by which LCR deformation in endothelial cells induces vascular structural changes at the molecular level, it is expected to pave the way for a turning point that will elevate the treatment outcomes of renal cell carcinoma.
๐ก These findings could bring about immediate changes in clinical practice regarding treatment decisions for patients with metastatic renal cell carcinoma. It becomes possible to perform diagnostic tests that pre-classify patients into high-risk and low-risk groups by analyzing the expression levels of the 15-gene panel in post-surgical tissue specimens or biopsy tissues. Based on the test results, the selection of first-line targeted therapies can be optimized. Even within the same TKI class, prioritizing sunitinib for high-risk patients and prescribing axitinib or pazopanib for low-risk patients can increase initial treatment response rates and minimize unnecessary exposure to drug toxicity. Furthermore, the newly identified CLCNKA target can be incorporated into new drug pipelines as a key target for developing next-generation targeted therapies and antibody-drug conjugates (ADCs) for high-risk patients who are refractory to existing TKIs and immunotherapy.
Chinese herbal medicines๐ค AI summaryDiscovery of 3 Key Pancreatic Cancer Oncogenes and Targeted Therapy Candidates via Multi-Cohort Transcriptome Network Analysis
Background Pancreatic ductal adenocarcinoma (PDAC) is a representative intractable malignancy with a 5-year relative survival rate of around 10%. Most cases are detected at advanced stages where surgery is impossible, and they rapidly develop resistance to standard chemotherapies like gemcitabine or FOLFIRINOX. The dense fibrous connective tissue matrix tightly surrounding cancer cells impedes drug penetration, resulting in treatment response rates remaining at the lowest levels. Despite large-scale cancer genome projects identifying major mutations like KRAS and TP53, the development of targeted therapies has faced a long period of stagnation. Experts point out that previous studies focused on single-mutation analysis struggled to capture the complex gene regulatory interaction networks within tumor cells. Furthermore, expression analyses based on small patient cohorts often failed during clinical validation due to high heterogeneity between cohorts. There was an urgent need for a multi-faceted computational biology framework to identify molecular targets and effective drugs that precisely target the core vulnerabilities of cancer cell survival networks. Key Findings The researchers developed a computational framework to construct large-scale co-expression networks by integrating transcriptomic big data from several independent PDAC clinical cohorts. After linking patient prognosis signature data, we precisely identified prognosis-associated functional modules through hypergeometric enrichment analysis. Subsequently, they applied multi-metric centrality analysis encompassing closeness centrality and betweenness centrality to identify the central regulatory factors within the network. Following strict filtering, 14 key genes that maintained high centrality across multiple cohorts were finalized. In the functional validation stage, the researchers employed a method of comprehensively cross-referencing single-cell RNA sequencing (scRNA-seq) and genome-wide CRISPR dependency screening data. scRNA-seq analysis revealed that among the candidates, three typesโCytoskeleton-associated protein 2-like (CKAP2L), Kinesin family member 18A (KIF18A), and Minichromosome Maintenance Complex Component 10 (MCM10)โwere specifically elevated in epithelial tumor cells rather than stromal cells. Clinical survival analysis of large patient cohorts clearly demonstrated that high expression of these three genes is directly linked to shortened patient survival. CRISPR loss-of-function screening data provided decisive evidence that these three genes are essential drivers sustaining the survival and division of PDAC cells. To drive therapeutic translation, a compound perturbation profile matching strategy using the Library of Integrated Network-based Cellular Signatures (LINCS) was employed. This algorithm ranks small-molecule compounds that revert PDAC-specific transcriptomic expression patterns to normal levels using a cross-cohort consensus scoring system. As a result, Mitoxantrone and OTSSP167 were selected as CKAP2L inhibition candidates, PF-3758309 as a KIF18A inhibition candidate, and Borinostat as an MCM10 inhibition candidate. In laboratory validations, including quantitative reverse transcription PCR (qRT-PCR), Western blot, and CCK-8 cell viability assays, these candidate drugs inhibited the mRNA and protein expression of target genes in a concentration-dependent manner and strongly suppressed cancer cell proliferation. Significance and Outlook This achievement is evaluated as having seamlessly connected the previously fragmented processes of target discovery and drug screening within a computational network platform. By combining multi-cohort cross-validation with single-cell transcriptomics and functional genomics data, the reproducibility of the identified targets was significantly enhanced. In particular, it provided a clue to target difficult-to-access cell division and DNA replication regulatory factors using approved drugs or compounds with proven efficacy. However, practical barriers to clinical implementation remain. Since validation has been primarily conducted using in vitro cell line models, it is pointed out that follow-up evaluations of drug penetration and in vivo pharmacokinetics are required in patient-derived organoids (PDO) and animal models that mimic the thick connective tissue characteristic of pancreatic cancer. Analysis suggests that optimizing dosage to minimize damage to normal pancreatic tissue and validating combination therapies with existing chemotherapies will be the keys to commercialization.
๐ก The established network-based discovery system can serve as a practical tool to significantly reduce the time and astronomical costs required for drug development. It provides direct guidance for drug repurposing strategies, such as the rapid repurposing of drugs with established safety profiles, like mitoxantrone or vorinostat, as treatments for refractory pancreatic cancer. In actual clinical practice, a companion diagnostic strategy could be established by testing the expression levels of CKAP2L, KIF18A, and MCM10 in patient tumor biopsies to select patient groups likely to respond to specific drugs. If a precision medicine scenario in which these inhibitors are selectively prescribed as combination therapies for refractory patients who do not respond to standard anticancer drugs becomes a reality, it is expected to mark a substantial turning point in improving patient survival rates.
Nature๐ค AI summaryLength variation in human genomic tandem repeats determines cell-specific gene regulation and complex trait variance
Background Human genome research has focused on identifying single nucleotide polymorphisms (SNPs) over the past decades. While Genome-Wide Association Studies (GWAS) have identified numerous disease-associated loci, a significant portion of the mechanism by which genetic predisposition manifests as actual disease remains unknown. The academic community refers to this as 'missing heritability' and has investigated its causes from various angles. A representative area that has been overlooked due to the limitations of existing analysis technologies is tandem repeats (TR). Tandem repeats are regions in which short nucleotide sequences are arranged consecutively, distributed widely across the genome. Because the sequences themselves repeat, it was difficult to accurately measure changes in the length of repeat units using past short-read sequencing techniques. In most variant studies, this region was often excluded as technical noise that causes analysis errors. Gene expression analysis methods that were limited to the tissue level also posed a limitation. This is because it is difficult to capture subtle transcriptional regulation changes occurring in specific immune cells or rare cell populations when individual cells performing different functions are mixed. To reveal the mechanisms of complex diseases, a new approach that organically connects structural variations in the genome with molecular phenotypes at the individual cell level was urgently needed. Key Findings Researchers demonstrated at a genome-wide scale the impact of tandem repeat length variation on human cell gene expression by combining high-depth whole-genome sequencing (WGS) data with single-cell transcriptome analysis techniques. The research team employed a strategy of precisely profiling the diversity of tandem repeat lengths appearing in human cell groups and directly comparing them with cell-type-specific gene expression measurements. As a result of the analysis, the difference in the unit length of tandem repeats appeared as a key factor inducing cell-type-specific expression disparities by interacting with adjacent regulatory regions. Unlike SNPs, which involve the substitution of a single base, physical structural changes resulting from the expansion or contraction of repeat units altered transcription factor binding affinity or locally regulated chromatin accessibility. It was revealed that even with the same tandem repeat length variation, different regulatory effects are induced depending on the immune cell type, such as B cells or T cells. Tandem repeat length variation also directly contributed to the formation of the genetic architecture of complex traits. Integrating tandem repeat length variation into existing SNP-based analysis models significantly enhanced the genetic explanatory power when explaining susceptibility to multifactorial diseases, including immune disorders. This demonstrates that beyond simple sequence differences, the physical repeat count within the genome acts as a powerful epigenetic and transcriptional regulatory switch that shapes the unique identity of cells and their disease vulnerability. Significance and Outlook This study represents a significant achievement in expanding the scope of the human genome map, which had been biased toward single-nucleotide variants, to include polymorphic repeat regions. By elucidating at the molecular level how structural variations induce cell-specific regulatory abnormalities in non-coding regions, it is expected to invigorate research into the pathogenesis of complex diseases. A foundation has been laid to improve the precision of algorithms for predicting personalized risk in diseases with significant inter-individual symptom variability, such as autoimmune or chronic inflammatory diseases. Identifying repeat sequence targets that cause abnormal gene overexpression specifically in certain immune cells could provide new regulatory targets for consideration during the development of precision therapeutics. Challenges remain to be solved before generalizing to clinical practice. To precisely decode complex repeat regions, long-read sequencing technology must become widespread and be accompanied by reduced testing costs. Subsequent research to establish a standard reference database defining the typical length of tandem repeats across diverse ethnic groups worldwide is necessary to ensure clinical validity.
๐ก This research provides practical tools to advance the precision molecular diagnosis and drug target discovery systems for complex immune diseases. For patients with autoimmune diseases such as rheumatoid arthritis and multiple sclerosis, in whom the cause could not be identified through existing gene panel tests, performing tandem repeat length measurement alongside testing can more precisely stratify individual disease risk and immune cell reactivity. Drug development companies can diversify their research pipelines by discovering antisense oligonucleotides (ASOs) and small molecules that block or regulate transcription factor binding hyperactivated due to abnormal repeat sequence lengths. Practical application scenarios are also expected, such as using tandem repeat length as a selection criterion to evaluate the stability and function of specific immune cells in the manufacturing process of cell therapies.
Nature๐ค AI summaryUncovering the Cellular Mechanisms of Immune Disease Genetic Variants via Single-Cell Transcriptome Analysis of 2,000 Individuals
Background For decades, genome-wide association studies (GWAS) have identified numerous genetic variants that increase the risk of disease. However, more than 90% of the discovered variants are located in non-coding DNA regions that do not directly specify protein amino acids. It was no small task to interpret how these variants lead to actual diseases through specific molecular pathways. This is because non-coding region variants primarily serve as switches that finely tune the expression levels of neighboring genes. Gene expression quantitative trait loci (eQTL) analysis has become a key methodology for identifying these regulatory switches. Past studies have largely relied on analyzing bulk tissue, where tens of thousands of cells are mixed together. In tissues where numerous immune cells are intertwined like blood, a fatal limitation is revealed. This is because significant expression changes occurring in rare cells within the blood were drowned out and lost amidst the signals of the majority of other cells. It is common for the same genetic variant to perform opposite functions depending on the cell type or to regulate gene expression only in specific immune cells. Previous bulk analysis methods could not isolate these cell-type-specific characteristics. To understand where individual genetic differences create waves within the immune system, large-scale analysis at single-cell resolution was essential. Key Findings An international research team, led by the Garvan Institute of Medical Research and the Murdoch Children's Research Institute in Australia, has released the results of the first phase of the 'TenK10K' project in the international journal Nature. The study involved precisely profiling the whole genome and single-cell transcriptomes of over 5 million individual cells isolated from the Peripheral Blood Mononuclear Cells (PBMC) of 1,967 donors. By integrating single-cell RNA sequencing (scRNA-seq) technology, the researchers classified immune cells in the blood into 28 distinct subtypes and rigorously analyzed the correlation between genetic variants and gene expression. Through this process, they achieved the feat of mapping risk loci associated with 16 complex immune diseases and 44 hematological phenotypes at the cellular level. As many as 1,500 disease candidate genes are attributed to specific cell subtypes. A notable achievement is the demonstration that the effects of disease-related genetic variants are observed exclusively during specific differentiation stages or activation states of certain immune cells. Variants that increase the risk of rheumatoid arthritis or systemic lupus erythematosus selectively disrupted gene expression only in specific micro-subtypes of helper T cells, rather than in all lymphocytes. Faint signals that were dismissed as statistical noise in past bulk tissue analyses transformed into clear biological pathways upon reaching single-cell resolution. This clearly demonstrates that dysregulation of micro-immune cells is the molecular mechanism determining disease susceptibility. Significance and Outlook The completed single-cell eQTL map serves as a robust stepping stone to narrow the gap between genomic information and actual pathogenesis. This is because it provides precise coordinates for the cellular components in the body that malfunction due to risk factors of complex genetic diseases. In new drug research, it is expected that the search for targets that selectively attack only the cell subtypes causing the disease will become much easier. This opens a path to reducing the side effects of existing immunotherapies that increase infection risk by indiscriminately suppressing systemic immunity. However, clear challenges remain. Since this data was primarily collected from blood cells of healthy individuals, it has limitations in fully representing the cellular states within chronic inflammatory tissues where disease is progressing. While the sample of approximately 2,000 people is a leap forward compared to previous single-cell studies, the statistical power is still lacking to capture the influence of all rare variants. Broadening the cohort to include diverse racial groups and integrating it with spatial transcriptomics technology, which captures cellular positional information within tissues, will be the key determinant of future completeness.
๐ก The pharmaceutical and biotech industries can use this map as an accelerator in the early stages of drug discovery. If it is known that a specific autoimmune disease variant causes overexpression of a target protein only in activated B cells or helper T cell subtypes, a strategy of designing small molecule compounds or antibody-drug conjugates (ADCs) that selectively target those cells from the outset is effective. This approach can reduce unnecessary immunosuppressive side effects and increase the success rate of clinical trials. In the field of clinical genomic diagnostics, it contributes to establishing personalized treatment strategies. When a disease-risk variant is discovered in a patient's whole-genome sequence, it becomes possible to practically implement personalized medicine by predicting which immune cell pathways will be primarily damaged and taking targeted preventive measures even before symptoms appear.
Nature Genetics๐ค AI summaryIdentifying Mechanisms of Cardiometabolic Disease Variant Expression via a Single-Nucleus Multi-Omics Map of Human Skeletal Muscle
Background Skeletal muscle is the largest metabolic organ in the human body, driving glucose metabolism and insulin response. Many genetic variants that increase the risk of cardiometabolic diseases, such as type 2 diabetes or dyslipidemia, are deeply intertwined with metabolic pathways in skeletal muscle. While Genome-Wide Association Studies (GWAS) have identified thousands of disease-risk loci, the molecular pathways through which non-coding variants lead to disease remain unclear. Previous transcriptome analyses and genomic studies have relied on bulk analysis methods that grind entire tissue lysates. In muscle tissue, which is a complex mixture of muscle fibers, satellite cells, fibro-adipogenic progenitors, endothelial cells, and immune cells, it was nearly impossible to detect subtle gene regulation abnormalities specific to each cell type or physiological state. The technical difficulty of applying conventional single-cell isolation techniques due to the structural uniqueness of muscle fibers as multinucleated cells also posed a barrier. There was an urgent need for a high-resolution molecular map to reveal in which cell-nuclear states disease-associated variants cause aberrant regulation. Key Findings Researchers constructed an integrated multi-omics reference map by simultaneously performing single-nucleus chromatin accessibility analysis (snATAC-seq) and single-nucleus transcriptome analysis (snRNA-seq) based on isolated single nuclei from human skeletal muscle tissue. After precisely isolating myofiber subtypes (Type I, Type IIa, Type IIx) and the stromal cell nuclei constituting the microenvironment, they compared chromatin accessibility and transcriptional activity at the single-nucleus level. Systematic mapping of single-nucleus expression quantitative trait loci (eQTL) and chromatin accessibility quantitative trait loci (caQTL) revealed a distinct group of regulatory variants that function only in specific cellular states and physiological contexts. When analyzing the entire tissue, thousands of cell-specific QTLs that were obscured by average signals came to light. By combining GWAS statistical signals for metabolic diseases with colocalization analysis of these QTL signals, the causal gene regulatory circuits underlying disease susceptibility were revealed. The mechanism by which specific risk alleles selectively block chromatin accessibility and distort gene expression at distal regulatory element sites involved in insulin signaling and glucose transporter (GLUT4) translocation pathways was directly confirmed. Significance and Outlook This study clearly demonstrates that non-coding genetic variants causing cardiometabolic diseases induce pathological transcriptional dysfunction only within specific cell types and functional states inside skeletal muscle. By narrowing the horizon of target discovery from the gene level to cell-state-specific regulatory regions, a foundation has been laid to significantly reduce failure rates in drug candidate selection. However, since the map was constructed primarily using static tissue biopsy samples, follow-up studies are needed to track how the chromatin environment is restructured during dynamic metabolic stimuli, such as exercise or acute dietary changes. Expanding the ethnic diversity of skeletal muscle biopsy samples and standardizing processes to lower the analysis costs of the single-nucleus multi-omics pipeline are also challenges to be addressed in the future.
๐ก The single-nucleus multi-omics regulatory map constructed in this study serves as an immediate guide in the target validation stage of drug development. When developing treatments for type 2 diabetes, pre-identifying that a specific candidate gene causes metabolic abnormalities selectively in Type I muscle fibers or fibro-adipogenic progenitors of skeletal muscleโrather than in all cells in the bodyโcan reduce on-target toxicity and maximize drug efficacy. In designing gene therapies or antisense oligonucleotides (ASOs) targeting non-coding regulatory regions, the exact chromatin-open sites containing disease-causing variants can be targeted, which is expected to accelerate the early realization of precision medicine for metabolic diseases based on patient genotypes.
Microorganisms๐ค AI summaryFrom Retrospective Reporting to Proactive Prediction: AI and Genomics Reshaping the Antimicrobial Resistance Defense Line
Background Antimicrobial Resistance (AMR) is a critical health threat undermining the foundations of modern medicine. Existing surveillance systems have remained reactive, relying on post-hoc reporting after culturing resistant bacteria from clinical settings and testing drug susceptibility. This often leads to missing the window for transmission prevention, as the reality of resistance is only identified after it has spread through communities or ecosystems. Overburdened healthcare systems and insufficient diagnostic infrastructure further widen the time lag in statistical aggregation. Ecosystem disturbances due to climate change, localized armed conflicts, and large-scale population movements are twisting infectious disease transmission paths in unpredictable directions. In low- and middle-income countries, deep statistical gaps persist due to a lack of laboratory equipment and specialized personnel. This is why critics point out that traditional culture-based epidemiological investigations cannot keep pace with the rapid mutation rate of surging multidrug-resistant bacteria. Even accumulated resistance information has been left fragmented across human, animal, and environmental sectors. A paradigm shift is urgently needed to capture warning signs in advance and deploy preemptive defenses. Key Findings This research presents a methodology to transition from a retrospective reporting system to proactive predictive intelligence by combining Artificial Intelligence (AI), Whole-Genome Sequencing (WGS), and metagenomics. Thanks to next-generation sequencing technologies that can read entire microbial communities, it is now possible to identify Antibiotic Resistance Genes (ARG) at the genomic level, even for pathogens that are difficult to culture. By integrating Machine Learning (ML) algorithms, researchers can precisely pinpoint unknown resistance mutations and potential transmission pathways by synthesizing sequence patterns and clinical information. In particular, the utility of wastewater-based genomic epidemiology in environmental surveillance has been proven. The study explains that analyzing wastewater treatment plant samples using metagenomic technology can capture trends in the spread of resistance genes weeks before clinical diagnoses. This provides a shield for early warnings at the population and ecosystem levels. However, current digital surveillance networks show limitations, remaining confined to individual sectors or relying on short-term project grants. A unified 'One Health' platform encompassing humans, animals, and the environment has not yet reached the operational stage anywhere in the world. The researchers also presented a blueprint for non-antibiotic therapeutic pipelines to overcome drug resistance. Bacteriophages, which specifically lyse target bacteria, and CRISPR-based antimicrobials, which precisely target resistance genes, are emerging at the forefront. Antimicrobial Peptides (AMP) that destroy bacterial cell membranes, microbiome normalization technologies, and nanomedicines are also listed among the next-generation weapons. Host-directed therapies, photodynamic therapy, and resistance-preventing vaccines are also identified as essential components of the core portfolio. Significance and Outlook The predictive platforms led by AI and genomic epidemiology are gaining attention as the key to bridging the speed gap between antibiotic development and public health response. Considering that developing new drugs takes over 10 years and massive costs, establishing early warning networks is an economical way to extend the effective lifespan of existing drugs. Non-antibiotic alternatives, such as phage therapy and CRISPR antimicrobists, are accelerating their entry into clinical trials as customized treatments targeting multidrug-resistant superbugs. The pharmaceutical industry also gains an opportunity to shift research directions toward targeted biotherapeutics, reducing the burden of developing high-failure-rate broad-spectrum antibiotics. However, significant hurdles remain before these research achievements lead to actual commercialization. High-cost sequencing equipment and data analysis computing resources are concentrated in some developed nations; if regional imbalances in data collection are not resolved, the predictive power of AI models may remain halved. Differences in international surveillance data sharing regulations and licensing standards are also identified as immediate challenges to be addressed. A practical quarantine network capable of confronting borderless microbial threats can only be completed if supported by multidisciplinary collaborative systems and stable public financial support.
๐ก The integrated surveillance system and non-antibiotic treatment models discussed in this study provide concrete application points for infection control departments in medical institutions and quarantine authorities. If wastewater pipeline monitoring in large hospitals is linked with genomic analysis, the influx or hospital-wide outbreaks of Carbapenem-Resistant Enterobacterales (CRE) in intensive care units can be rapidly detected even before clinical specimen confirmation. If AI mutation prediction algorithms are introduced to diagnostic laboratories, the time required for drug susceptibility testingโwhich typically takes more than two daysโcan be reduced to within a few hours, dramatically increasing the accuracy of initial prescriptions. In the field of drug development, this serves as a guideline to increase the clinical success rate of non-antibiotic therapeutic pipelines. When designing synthetic antimicrobial peptides or CRISPR delivery vehicles, it becomes possible to implement strategies that preemptively block the potential for target bacteria to evade mutations by cross-verifying with wastewater and clinical genomic databases.
International journal of pharmaceutics๐ค AI summaryChondrocyte Membrane-Camouflaged Nanoparticles Designed via Machine Learning Induce Cartilage Regeneration in Degenerative Osteoarthritis
Background Osteoarthritis (OA) is a representative degenerative joint disease affecting hundreds of millions of elderly people worldwide. The destruction of the cartilage matrix that protects joints, coupled with persistent chronic inflammatory responses, leads to irreversible joint damage. Because cartilage tissue lacks blood vessels or lymphatic vessels, it lacks self-regenerative capacity. Existing clinical practices have relied on symptomatic therapies, such as non-steroidal anti-inflammatory drugs or intra-articular steroid injections, to alleviate pain. However, these drug administrations fail to fundamentally repair the damaged cartilage matrix and are limited by the rapid clearance of drugs injected into the joint cavity via synovial circulation. The academic community has focused on messenger RNA (mRNA) delivery technology to induce the expression of SOX9, a transcription factor that promotes cartilage differentiation and matrix formation. However, safely delivering mRNA through the chondrocyte surface into the cytoplasm is not easy. While lipid nanoparticles (LNPs) are a representative gene carrier, precise optimization of the formulation composition was essential to evade immune responses within the joint cavity and penetrate the dense cartilage extracellular matrix. The conventional approach of experimentally verifying lipid component ratios or electrical properties one by one has been criticized as an obstacle to clinical entry due to its high time and cost consumption. Key Findings The researchers overcame these limitations by establishing a CM-LNP@mRNA platform that fuses a Random Forest (RF)-based machine learning algorithm with chondrocyte membrane coating technology. As a result of screening candidate combinations of lipid formulations using machine learning, they identified the nitrogen-to-phosphorus (N/P) ratio and the ratio of the ionizable lipid SM-102 as key design variables. The feature importance calculated from the RF analysis showed that the N/P ratio was 0.57 and the SM-102 ratio was 0.32, statistically proving that these two factors govern encapsulation efficiency and inlet hydrodynamics. When chondrocyte membranes were coated onto the optimal formulation derived from machine learning, the physicochemical properties of the carrier significantly improved. The encapsulation efficiency of the final completed CM-LNP@mRNA formulation reached 95.50%, and the average particle size was controlled at 133.70 nm. This stands in stark contrast to the control formulation prepared using conventional methods under the same conditions, which showed an 89.20% encapsulation rate and a size of 145.40 nm. Biophysically, it was confirmed that the surface zeta potential shifted from -25.57 mV to -35.07 mV after the cell membrane coating process, indicating that the cell membrane components were stably anchored on the particle surface. Efficacy verification was performed in a chondrocyte model where an inflammatory environment was induced with interleukin-1 beta (IL-1ฮฒ). When cells were treated with CM-LNP@mRNA, the relative expression of Col2a1, a key component of the cartilage matrix, increased from 0.20 to 0.33, and Acan expression was restored from 0.50 to 0.73. Conversely, the expression of matrix metalloproteinase-3 (MMP3), which degrades cartilage, plummeted from 10.50 to 7.80, and ADAMTS5 expression dropped from 4.00 to 2.50. These changes at the transcriptomic level led to the inhibition of protein secretion. Measurements at day 7 of administration showed that the concentration of the degrading enzyme MMP-3 dropped to half, from 37.10 ng/mL to 18.50 ng/mL. The concentration of prostaglandin E2 (PGE2), a major inflammatory mediator, also significantly decreased from 113.70 pg/mL to 76.50 pg/mL. In rat models of degenerative arthritis, the volume of osteophytes (abnormal bone growth tissue within the joint) significantly decreased from 4.00 mmยณ, inhibiting structural joint degeneration. Significance and Outlook This study is significant in that it utilized an AI-based predictive model to rapidly derive complex nano-formulation conditions and increased tissue targeting through biomimetic cell membrane technology. The method of coating synthetic nanoparticles with homologous chondrocyte membranes provides an immune camouflage effect that avoids phagocytosis by macrophages in the synovial fluid and maximizes affinity with the cartilage extracellular matrix. It is regarded as a strategy that simultaneously addresses the immunogenicity issues of existing viral vector carriers and the limitations of non-specific distribution in lipid-based delivery systems. Challenges remain to be solved before achieving clinical application. Since human articular cartilage is much thicker and the synovial fluid volume is larger compared to rodent models, verification through large animal models is required to determine whether the same tissue penetration and persistence are observed in the human joint environment. The standardization of the cell membrane extraction process, ensuring batch-to-batch uniformity during mass production, and evaluating potential immune reactivity caused by cell membrane-derived proteins are also identified as tasks for follow-up research.
๐ก This research presents a specific formulation platform that can shift osteoarthritis treatments from simple pain management to Disease-Modifying Osteoarthritis Drugs (DMOADs). In clinical settings, it can be administered via direct intra-articular injection, promoting direct SOX9 protein synthesis in damaged chondrocytes without the burden of systemic side effects in elderly patients. It is expected to function as a therapeutic option for patients with moderate-stage arthritis awaiting joint replacement surgery, by slowing the rate of cartilage degeneration and inducing autologous cartilage regeneration. Industrially, it provides a commercializable pipeline model that can drastically reduce the cost of developing drug delivery systems by shortening the search period for complex geological nanoparticle candidates from months to days using artificial intelligence.
Nature Genetics๐ค AI summaryDifferences in Transcriptome Annotation and RNA Quantification Methods Reverse eQTL and TWAS Disease Gene Analysis Results
Background Genome-wide association studies (GWAS) have expanded the horizons of modern medicine by discovering numerous genetic variants associated with diseases. However, a major challenge remains: most disease variants are located in non-coding regions that do not directly encode proteins, making it difficult to reveal specific biological mechanisms. To fill this gap, the academic community introduced expression quantitative trait loci (eQTL) analysis. This is an approach that statistically identifies the impact of genetic variation on tissue-specific gene expression levels. Furthermore, transcriptome-wide association study (TWAS) and colocalization analysis, which combine eQTL reference panels with GWAS summary statistics, have become key tools for identifying causal genes driving disease onset. The problem lies in the foundation that guarantees the reliability of the analysis. While the academic community has applied strict standards to complex post-hoc statistical modeling and multiple testing corrections, it has been relatively lax regarding the upstream preprocessing stages of ribonucleic acid sequencing (RNA sequencing, RNA-seq), which serve as the starting point for analysis. It has overlooked the fact that downstream research results can be destabilized depending on which transcriptome annotation database is used to convert raw sequence reads into gene expression levels, and whether alignment-based or pseudo-alignment-based quantification is chosen. Key Findings An international research team demonstrated in a paper published on October 26 in the journal Nature Genetics that the choice of upstream RNA-seq processing steps can significantly impact genomic analysis. The team systematically compared combinations of transcriptome annotation systems and expression quantification pipelines using identical genomic and transcriptomic raw data. Clear discrepancies emerge starting from the baseline setting of transcriptome annotation. When applying different annotation versions with varying GENCODE, Ensembl, or exon definition ranges, the assigned gene-specific expression levels for the same sequencing reads changed significantly. In particular, quantification distortions were prominent in genomic regions where multiple transcript isoforms coexist or gene boundaries overlap. These preprocessing biases are directly propagated to the eQTL mapping stage. The composition of statistically significant eQTL variant-gene pairs fluctuated significantly due to differences in quantification algorithms alone. The concordance rate of eQTL discovery between traditional alignment-based quantification tools and algorithms applying transcript deconvolution fell below expectations. Downstream analyses, such as TWAS and colocalization, are also directly affected. It was observed that the ranking of candidate genes identified as key causal mediators of disease risk reversed simply by changing the upstream preprocessing method. Even the posterior probability of colocalization, which evaluates whether GWAS signals and tissue-specific expression signals originate from the same variant, reached opposite conclusions depending on the preprocessing choice. The initial choice of data processing essentially dictated the final results of disease gene identification. Implications and Outlook This study issues a stark warning to the functional genomics research ecosystem. It demonstrates that even when advanced statistical models are applied, subtle biases introduced during the raw data preprocessing stage can distort the overall conclusions of the research. The researchers urged the establishment of specific methodological reporting standards to ensure data reproducibility in future genomic studies. They pointed out that researchers must move away from the practice of simply listing analysis pipeline names and instead specify in detail the build version of the transcriptome annotation used and the quantification parameters. A sensitivity verification procedure, which compares multiple preprocessing combinations rather than relying on a single pipeline, was also identified as an essential element. In the long term, the academic community anticipates that the widespread adoption of long-read sequencing technology will serve as a turning point in mitigating annotation incompleteness. This is because it can fundamentally reduce errors in isoform protein quantification by directly decoding the entire transcriptome without fragmentation. Attention is now focused on whether discussions regarding the standardization of algorithms and reference databases will intensify.
๐ก This is expected to mark a turning point in significantly reducing the failure rate of target identification in new drug development. Before investing massive clinical costs, biopharmaceutical companies use public eQTL and TWAS data to validate candidate targets. If an incorrect gene is selected as a target without recognizing the distortions caused by initial RNA-seq quantification methods, there is a high risk of wasting enormous capital and time in the preclinical stage. This study provides the basis for incorporating preprocessing sensitivity evaluation as an essential quality control (QC) item in drug candidate discovery pipelines. Only robust genes that maintain significance consistently across various transcriptome annotations and quantification conditions should be selected to increase the success rate of target efficacy validation. Genomic analysis service companies and the diagnostics industry are also poised to secure the legal and clinical reliability of their analysis results through standardized certification of data processing pipelines.
Nature Genetics๐ค AI summaryHigh-speed computational model fastGxE discovers hidden gene-environment interactions in large-scale biobanks
Background Genome-Wide Association Studies (GWAS), which explore the association between Single Nucleotide Polymorphisms (SNPs) and diseases, have served as a key tool for uncovering the genetic basis of complex diseases for decades. However, human biological responses are not determined by genetic information alone. Even among individuals with the same genetic variants, disease risk and biomarker levels vary significantly depending on external environmental exposures such as diet, exercise, alcohol consumption, and air pollution. This is why studying Genotype-Environment Interaction (GxE), where genotypes and environmental factors combine to alter phenotypes, has emerged as an essential task for precision medicine. Until now, GxE analysis has faced difficulties in scaling to large-scale studies due to computational limitations. Statistically testing every combination of tens of millions of genetic variants and dozens of environmental factors across hundreds of thousands of genomes requires massive computational power. The existing standard Linear Mixed Model (LMM) required repetitive matrix calculations every time an interaction term between a variant and an environmental factor was estimated. In modern biobank environments where sample sizes exceed tens of thousands and reach 500,000, completing analyses takes months and incurs enormous computing costs, leading researchers to abandon genome-wide exhaustive testing in favor of selecting and validating only a small number of targeted variants. Key Findings Researchers have presented fastGxE, a scalable algorithm that resolves computational bottlenecks and increases statistical power in large-scale biobank datasets. fastGxE redesigned the computational structure to process calculations in near-linear time, even with sample sizes of hundreds of thousands. By shifting from a process of repetitive matrix decomposition for each variant to a method of pre-calculating common statistics, unnecessary redundant calculations were eliminated. As a result, the analysis speed increased by dozens to hundreds of times compared to existing analysis pipelines, enabling genome-wide interaction scans even in standard single-server environments. Researchers applied this algorithm to the genomic and clinical data of approximately 500,000 individuals in the UK Biobank (UKB) cohort. They completed genome-wide GxE scans for physical measurements such as height and Body Mass Index (BMI), as well as approximately 30 types of blood biomarkers routinely measured in health checkups. As a result of the analysis, numerous previously unidentified interaction loci were discoveredโloci that did not exceed statistical significance in traditional GWAS due to weak main effects. Certain genetic variants related to lipid metabolism showed a significant effect in lowering blood triglycerides only in groups practicing regular aerobic exercise. Variants associated with liver enzyme levels also clearly demonstrated context-dependent genetic effects, in which the magnitude of expression was differentially regulated depending on smoking status or alcohol consumption. Significance and Outlook The emergence of fastGxE represents a technological leap, expanding the perspective of genomic research from analyzing the effects of fixed genetic variants to tracking dynamic interactions with the environment. Polygenic Risk Scores (PRS), widely used for disease risk prediction, have had accuracy limits because they sum only the average effects of genetic variants while excluding environmental factors. By integrating the gene-environment interaction terms identified by fastGxE, it becomes possible to calculate precise risk levels that reflect an individual's actual lifestyle. The task is also clear. Since this study was primarily conducted based on UKB data centered on European ancestry, additional verification through multi-ethnic biobanks is required to see if the same interactions are reproduced in diverse ethnic groups, such as Asian or African populations. A follow-up task involves advancing analysis models to complement the self-reporting errors inherent in survey-based environmental data by linking them with continuous biological signals collected from wearable devices like smartwatches.
๐ก In clinical settings, targeted lifestyle modification guidelines based on an individual patient's genetic vulnerability can be established. For example, even among patients with the same cardiovascular disease risk variants, a customized exercise prescription can be prescribed by identifying the subgroup that responds particularly sensitively to exercise therapy. It is also highly valuable for drug development and the pharmaceutical industry. During clinical trials, pre-stratifying patient subgroups with significant variability in drug responsiveness or metabolite concentrations due to environmental exposure differences via gene-environment (GxE) interaction analysis allows for the precise identification of patient populations to maximize drug efficacy and reduce clinical trial dropout rates.
Nature Genetics๐ค AI summaryGene Expression Signals Obscured by Linkage Disequilibrium: Uncovering Directional Distortion in Transcriptome-Wide Association Analysis
Background As human genome analysis technologies advance, research to identify target genes that cause diseases has been actively pursued. Genome-Wide Association Studies (GWAS) have succeeded in discovering numerous disease-related loci. However, because most variants are located in non-coding regions that do not encode proteins, it has been difficult to identify the causal genes that directly cause disease. To overcome these limitations, Transcriptome-Wide Association Studies (TWAS) have become a key tool in statistical genetics. TWAS statistically models the correlation between gene expression levels and disease phenotypes by integrating expression Quantitative Trait Loci (eQTL) data. Researchers have used this to estimate genes that increase or decrease disease risk in specific tissues or cells. Existing analysis pipelines possess a structural vulnerability: they fail to completely exclude the phenomenon of Linkage Disequilibrium (LD), where adjacent genetic variants are inherited together. This is because co-regulation occurs, where the expression profiles of surrounding genes change together due to linkage with nearby marker variants rather than the causal variant itself. Concerns have been steadily raised that false positive signals might be mistaken for disease causes or that the direction of causality might be reversed. Key Findings Researchers mathematically modeled and quantified the confounding effect of LD-driven co-regulation of expression profiles on TWAS results across cell types. They precisely disentangled the genetic structure shared by target genes harboring causal variants and nearby non-causal genes, applying it to both simulations and real genomic data. Analysis revealed that co-regulation entangled with nearby tagging variants goes beyond simple statistical noise, fundamentally altering the direction of gene-disease associations. A reversal phenomenon was systematically observed, in which genes that actually suppress disease occurrence were incorrectly estimated as risk factors promoting disease, obscured by strong linkage disequilibrium (LD) signals at adjacent loci. As expression data across tissues and cell types became more complexly entangled, the probability of such signal interference significantly increased. Particularly in complex polygenic diseases where multiple genes are clustered in a narrow genomic region, it was numerically proven that a non-negligible proportion of false correlation coefficients derived from co-regulation disturbance pass through existing correction models. The researchers clearly demonstrated, using statistical metrics, the limitation of existing TWAS algorithms that fail to incorporate both cell-specific transcriptional regulatory networks and local LD structures. While previous methodologies focused primarily on expanding sample sizes or improving the predictive power of machine learning models, this study differentiates its approach by presenting a quantitative metric to separately measure the multi-pleiotropic signal entanglement caused by associated variants. Significance and Outlook This analysis challenges the prevailing dogma in transcriptome research regarding methods for prioritizing disease candidate genes. In the early stages of discovering numerous drug candidates, there have been many cases where development failed in later phases because the directionality suggested by TWAS was blindly trusted, leading to the development of therapeutics that inhibit or promote the functions of target genes. If the direction of disease association was reversed due to statistical illusion, massive resources and time would inevitably be wasted in the preclinical verification process. The field of genomic epidemiology now faces the challenge of introducing a standard validation framework that verifies local variant structures alongside, rather than accepting transcriptome-wide association study (TWAS) results as the sole truth. It is essential to include a stage of physically validating statistical correlations by integrating high-resolution single-cell transcriptome data with functional genomics experiments based on CRISPR gene editing. At the same time, existing multi-ancestry genomic reference panels remain limited in their ability to fully capture the complexity of tissue-specific eQTLs and differences in linkage disequilibrium (LD) structures across ancestries. Follow-up research must focus on significantly expanding data from non-European populations and developing multivariate statistical correction algorithms that integrate the transcriptional variability of the cellular microenvironment.
๐ก It provides direct guidance to biopharmaceutical companies verifying the validity of disease targets in the early stages of drug development. It can be applied as a screening filter to pre-emptively filter out interference from linkage disequilibrium signals with adjacent genes, moving away from the practice of relying on top TWAS rankings in disease gene target discovery pipelines. For example, when developing antisense oligonucleotides or small-molecule compounds to lower the expression of a specific protein as a therapeutic target for autoimmune or metabolic diseases, pre-determining whether the gene is the actual causal target or merely reflects the signal of a linked variant can reduce the probability of failure in the development pipeline. Genomic analysis platform companies can increase analysis accuracy through software updates that add directionality reliability scores to existing association metrics.
Nature Genetics๐ค AI summaryDecoding 98% of the Genomic Dark Matter with Deep Learning: Current Status and Challenges of Noncoding Variant Prediction Models
Background After the completion of the Human Genome Project, the greatest barrier faced by molecular biologists was the functional interpretation of the noncoding region, which accounts for 98% of the genome. The exon regions, which encode proteins, allow for relatively easy prediction of variant pathogenicity by directly comparing amino acid sequence changes. On the other hand, the interpretation system for promoters, enhancers, cleavage junction sites, and non-coding RNA regions responsible for transcriptional regulation was extremely opaque, despite the fact that a single base change could drastically alter disease susceptibility. More than 90% of single nucleotide polymorphisms (SNPs) reported to be associated with diseases through Genome-Wide Association Studies (GWAS) are concentrated in these noncoding regions. However, existing bioinformatics tools have shown limitations in inferring human-specific regulatory variants or cell-line-specific epigenomic changes due to excessive reliance on multi-species sequence conservation scores. Although high-throughput mass screening experiments such as Massively Parallel Reporter Assays (MPRA) have been developed, verifying hundreds of millions of human variant combinations is accompanied by astronomical costs and technical constraints. This created an urgent need for computational models capable of quantifying the impact of noncoding variants on gene expression and chromatin structure using only sequence information. Key Findings In recent years, genomic deep learning technology has evolved beyond simple local Convolutional Neural Network (CNN)-based models into neural network architectures that capture long-range interactions spanning tens to hundreds of thousands of base pairs. If early DeepSEA or Basset models operated at the level of determining transcription factor binding or DNA nuclease sensitivity within sequences of a few thousand base pairs, recent models like Enformer and Borzoi combine Transformer architectures to compute even three-dimensional physical contacts between promoters and enhancers occurring over distances of several hundred thousand base pairs. was included in the scope. These advanced models predict CAGE (Cap Analysis of Gene Expression) signals, RNA transcription levels, and histone modification (ChIP-seq) patterns for human cell lines with single-nucleotide accuracy. We have reached a stage where in silico mutagenesis can simulate how much the expression of downstream target genes increases or decreases when an artificial mutation is introduced at a specific noncoding position. The Sei model, which classifies noncoding variants into dozens of sequence classes based on chromatin state to prioritize pathogenic variants in a disease-specific manner, has also yielded such results. However, the asymmetry of the model structure did not directly lead to full prediction accuracy. When evaluating rare variants, limitations remain in overfitting to the reference genome sequence, which leads to underestimating allele-specific expression differences, as well as in the ability to isolate subtle regulatory effects that occur only in specific tissues. The bias in the functional genomics datasets used for training is also cited as a key factor causing performance gaps between tissues. Significance and Outlook Genomic deep learning is forming a major turning point in elevating noncoding variants into the realms of precision medicine and drug target discovery. This provides the foundation to dramatically improve the diagnosis rate of rare, intractable genetic diseases, which previously remained as 'Variants of Uncertain Significance (VUS)' because their functional meaning could not be identified even as genomic decoding data from tens of thousands of patients accumulated. In cancer genomics, it is also beginning to serve as a key tool for uncovering the mechanisms of abnormal overexpression of oncogenes caused by somatic noncoding mutations. The future challenge is to bridge the gap between computer simulation results and actual in vivo molecular dynamics. It is required to build Genomic Foundation Models that reflect genetic polymorphism across diverse ethnic groups and to develop training techniques that directly incorporate single-cell transcriptome and multi-omics empirical data into the loss function. To ensure reliability in clinical settings, model interpretability must be enhanced to clearly trace back the specific patterns of transcription factor motif disruption that cause diseases.
๐ก The noncoding variant prediction models highlighted in this study can yield immediate cost-reduction effects in the target discovery and drug safety verification stages of drug development. By pre-screening enhancer variants around disease-associated GWAS genes using these models, regulatory regions that precisely control target gene expression can be selected computationally, reducing the need for unnecessary in vitro experiments. They also play a decisive role when designing antisense oligonucleotide (ASO) or CRISPR-based gene editing therapeutics. By precisely simulating potential off-target effects and unintended disruption of heterologous gene expression that may occur when targeting non-coding regulatory regions in advance, the safety profile of therapeutics can be established during the early stages of development.
Nature Genetics๐ค AI summary60 Genetic Variants of Liver Triglycerides Identified in a 70,000-Person Biobank Reveal Pathways to Liver Cancer Development
Background Non-alcoholic fatty liver disease (NAFLD) is a chronic metabolic disease that begins with simple fat accumulation and progresses to steatohepatitis (inflammation), fibrosis, and eventually cirrhosis and hepatocellular carcinoma (HCC). While affecting over 25% of the global population, the rate of progression and severity varies greatly among individuals. Even with similar alcohol intake or dietary habits, some remain at the simple steatosis stage, while others rapidly progress to malignant tumors. In clinical settings, efforts have been made to find molecular genetic clues that explain these deviations. Past studies have also reported representative genes involved in hepatic fat accumulation, such as PNPLA3 and TM6SF2. However, it has been difficult to explain the complex phenotypic differences across entire populations with just a few of these variants. Previous cohort studies were limited to sample sizes of a few thousand and struggled to capture subtle genetic signals due to errors in measurement methods for quantifying liver fat, such as liver biopsy or ultrasound. Intracellular triglyceride deposition in hepatocytes is the first trigger for disease progression. Therefore, a high-resolution genetic map capable of quantifying disease incidence by linking triglyceride levelsโmeasured via precise large-scale imagingโwith genomic information was urgently needed. Key Findings Researchers compared whole-genome sequences with abdominal MRI-based proton density fat fraction (PDFF) data from 71,488 UK Biobank participants using genome-wide association study (GWAS) techniques. The precise imaging measurements enabled quantification of intrahepatic triglyceride content without liver tissue damage. The analysis identified 60 genomic loci independently associated with liver triglyceride accumulation. Notably, about 30 of these are new signals not previously captured in liver disease research. Based on these discovered variants, the researchers reconstructed key regulatory pathways involved in lipid metabolism, insulin signaling, and intracellular lipid droplet formation. The research team constructed individual Polygenic Risk Scores (PGS) by converting the effect sizes of the 60 identified independent variants into weights. This score was not merely a predictor of simple liver fat volume. When linked with independent large-scale electronic health record (EHR) follow-up data, the top 10% high-risk PGS group showed a more than 3-fold increased risk of cirrhosis and a more than 2.5-fold increased risk of hepatocellular carcinoma compared to the bottom 10%. This provides numerical verification of the genetic trajectory in which fat accumulation in the liver parenchyma leads to long-term malignant diseases. Significance and Outlook This study expands liver metabolic disease research from a single-gene focus to a polygenic accumulation model. The 60 identified variants include protein groups involved not only in geological synthesis pathways but also in hepatic stellate cell activation and the creation of an inflammatory microenvironment. This provides a foundation for accelerating the discovery of new therapeutic targets. The development of RNA therapeutics or small molecule compounds that directly regulate the mechanisms of intrahepatic lipid accumulation, which were previously difficult to target, is expected to intensify. Significant changes are also anticipated in screening systems for the early detection of high-risk groups with asymptomatic liver cancer progression. Even if standard liver function blood markers (AST/ALT) fall within the normal range, individuals with a high PGS may be prioritized for regular detailed ultrasound or MRI screenings. However, the fact that this study's subjects are predominantly of European descent remains a challenge to address. Asian and African populations tend to exhibit different patterns of visceral fat accumulation relative to body mass index (BMI) and the incidence of liver fibrosis. Universal implementation of clinical practice guidelines will be realized only with subsequent validation in multi-ethnic cohorts and supporting functional genomic cellular experiments.
๐ก The polygenic risk score developed in this study can be directly implemented as a personalized precision risk classification tool in primary care for liver disease. Typically, liver diseases show no distinct symptoms until they reach the stages of end-stage cirrhosis or liver cancer. If high-risk groups are identified via a one-time genomic chip analysis during regular health checkups, preventive interventions such as early dietary/exercise therapy and preemptive prescription of anti-fibrotic drugs can be administered. This can also shift the paradigm of clinical trials for drug development. A major reason for the high failure rate of clinical trials for Metabolic Dysfunction-Associated Steatohepatitis (MASH) treatments has been the extreme heterogeneity in the rate of disease progression among patients. By selecting only patient groups with a high genetic risk of liver cancer for Phase 2 and 3 trials, the efficacy of new drugs can be statistically proven more quickly with fewer patients, significantly reducing development costs and duration.
Nature Genetics๐ค AI summaryNew metric tracking evolutionary constraints in 462 mammal species improves selection of rare genetic disease causative genes
Background With advancements in next-generation sequencing technologies, the cost of decoding patient exomes or whole genomes has significantly decreased. However, accurately identifying the actual causative variants that induce diseases from genomic data remains a difficult challenge. Existing metrics built on human population genomic data have evaluated gene importance by measuring how much loss-of-function variants are depleted within a population. Representative examples include gnomAD's probability of loss-of-function intolerance (pLI) and the upper bound of the observed-to-expected ratio (LOEUF). While these human population-based metrics have contributed to identifying genes for fatal severe diseases, they possess fundamental limitations. Due to the statistical limitations of sample sizes in human population data, it was difficult to statistically determine whether variant depletion was significant in genes with short sequence lengths. Genes subject to weak natural selection pressure, such as those associated with recessive genetic disorders or diseases that manifest in adulthood, were frequently omitted from analysis because the signal of variant depletion within the population was not distinct. Furthermore, there was a lack of tools to precisely evaluate the harmfulness of missense mutationsโwhere a single amino acid changesโon protein function at the gene level. Key Findings The researchers expanded their focus from the population genomics of a single human species to all 462 mammal species. This was the background for developing GISMO and GISMO-mis, new metrics that measure gene essentiality by quantifying the purifying selection signals preserved across the mammalian evolutionary trajectory spanning approximately 100 million years. Based on multiple sequence alignment data of 462 mammal species, the research team quantified how strictly each gene-coding region has excluded mutations during the evolutionary process. GISMO evaluates the intensity of evolutionary constraints across a gene, while GISMO-mis calculates a constraint index specialized for deleterious missense mutations that have significant impacts on protein structure and function. Analysis results showed that GISMO and GISMO-mis form a complementary relationship with existing human population-based metrics, significantly improving the discriminative power for gene deficiency. In more than 1,000 short protein-coding genes where it was difficult to distinguish essential genes using gnomAD metrics, GISMO captured distinct evolutionary conservation signals and accurately classified them as essential genes. This demonstrates that even if loss-of-function variants are rarely observed in human populations, regions with amino acid sequences strictly conserved throughout mammalian evolutionary history carry a high risk of disease. The utility of the new metrics was also confirmed in the analysis of an actual cohort of patients with rare developmental disorders. When the research team re-analyzed data from neurodevelopmental disorder patients carrying de novo mutations using GISMO-mis, candidate causative genes that had remained low in priority with existing metrics were repositioned to the top. In particular, the accuracy of predicting the pathogenicity of missense mutations related to congenital heart defects and autism spectrum disorder showed a significant improvement compared to previous models. Significance and Outlook This study clearly demonstrates with robust data that the big data of comparative genomics can solve long-standing challenges in human genomics medicine. The 100-million-year natural filtration mechanism provided by the evolutionary data of 462 mammal species precisely fills the gaps that databases of hundreds of thousands of humans could not fill. It has opened a new analytical paradigm for discovering disease genes by integrating human population data with evolutionary conservation data. Challenges also remain. While metrics based on interspecies comparisons are excellent for identifying fundamental biological mechanisms shared across all mammals, it is difficult to exclude the possibility that genes related to immune responses or higher brain functions uniquely developed in humans may be measured as having low evolutionary constraint. Follow-up research expanding the model to non-translated regulatory regions or non-coding RNA mutations is also required. Despite these constraints, if the GISMO metric group is incorporated into the pipeline for prioritizing disease candidate genes, it is expected to become a key asset in reducing the undiagnosed rate in the field of clinical genomic diagnosis.
๐ก More than half of patients suspected of having rare diseases who request exome analysis in clinical settings remain undiagnosed because the causative variant cannot be identified. This is because many variants of uncertain significance (VUS), whose functions are not clearly understood, are detected. GISMO and GISMO-mis can be directly utilized as filters in the diagnostic pipeline to rapidly identify actual pathogenic variants among numerous VUS. In the drug development industry, it also significantly reduces trial and error in the target discovery stage. If a specific gene shows a high GISMO score across mammals, the risk of inducing severe toxicity upon inhibiting that gene can be predicted in advance. Conversely, if a gene that appeared to have low importance in human population data shows high evolutionary constraint in diseased tissue, strategic decision-making becomes possible to prioritize it as a strong therapeutic target.