Background

Metagenomics has revolutionized the study of microbial ecosystems by enabling the analysis of genetic information from environmental samples without the need for cultivation. However, the rapid increase in data generated by high-throughput genome sequencing has placed a significant burden on existing information processing methods. Traditional alignment-based methods, which compare sequences pairwise, suffer from a critical limitation: their computational cost increases exponentially with the number of sequences being analyzed. Comparing hundreds of thousands of microbial genomes requires astronomical computational resources. Furthermore, each time a new genome is discovered and added to the existing database, the entire map must be redrawn, which is cumbersome. Researchers have also pointed out that the existing bacterial classification criteria cannot be directly applied to viruses, which evolve rapidly and exhibit high levels of variation. There is a pressing need for an analytical framework that can rapidly classify large-scale genomic information and be flexibly updated.

Key Findings

The research team has developed MetaSBT, a metagenomic framework based on a sequence Bloom tree (SBT), which integrates k-mer information (a sequence of a specific length) and a probabilistic data structure called a Bloom filter. This framework is divided into three modular stages: index construction, classification assignment, and dynamic updating, which improves computational efficiency. First, 26,285 viral reference genomes registered in the National Center for Biotechnology Information (NCBI) GenBank are collected to construct a basic index. Subsequently, 1,111 metagenome-assembled genomes (MAGs) and approximately 190,000 genomes from the Metagenomic Gut Virus (MGV) database are sequentially input to update the database.

The analysis revealed that MetaSBT identified a total of 40,729 species-level clusters. Interestingly, only about 20% of these clusters matched the existing database. The remaining 80%, or 32,560 clusters, were classified as novel candidate species not previously reported in the scientific literature. To improve the reliability of the analysis, a filtering process was also performed to refine the genomes based on an average nucleotide identity (ANI) of 99%. This process clearly identified 302 high-confidence novel viral species, removing redundant sequences and ensuring that each cluster contained at least three distinct genomes. In terms of computational speed and scalability, MetaSBT demonstrated remarkable improvements. The existing alignment-based tool, VIRIDIC, requires one billion operations to compare 30,000 or more genomes and cannot even be executed. In contrast, MetaSBT demonstrated linear scalability, with resource usage increasing proportionally as the genome analysis scale increased from 1,000 to 100,000.

Significance and Prospects

This research is considered to have laid the foundation for researchers worldwide to precisely classify and compare metagenomic data. The MetaSBT database can be quickly integrated with classification programs such as Kraken2, leading to a qualitative leap in microbial community analysis. In fact, it has also performed a cleansing function by correcting classification errors in existing viral information accumulated in public databases. The analysis revealed that 686 reference genomes stored in the NCBI GenBank and 1,051 viral metagenome-assembled genomes (vMAGs) contained incorrectly annotated information, which was subsequently corrected.

However, the computational structural limitations remain to be addressed. Due to the nature of the Bloom filter tree, the k-mer size specified at the beginning of the analysis cannot be arbitrarily changed. Changing the variable requires regenerating the database from scratch. The actual function and infection mechanism of the 30,000 or so novel candidate viruses identified by computer analysis in human or environmental settings have not yet been elucidated. Only when wet-lab experiments are conducted to analyze their specific functions and infection mechanisms, in addition to acquiring sequence information, will the full value of the tool be demonstrated.