Artificial intelligence enhances the precision of gene editing: OptiPrime, a predictive model for prime editing, emerges.
Background
Prime editing (PE) is a third-generation gene editing technology that precisely modifies the genomic DNA sequence. Unlike conventional gene editing tools, it can insert or delete desired sequences without completely cleaving the DNA double strand, making it a safe therapeutic tool. However, finding the optimal sequence for the prime editing guide RNA (pegRNA) that targets the desired editing site remains challenging. Researchers must experimentally verify the editing efficiency, which varies depending on the microenvironment surrounding the target sequence. It is difficult to intuitively select the sequence with the highest efficiency among numerous pegRNA candidates.
In the gene editing process, the mismatch repair (MMR) system, an intrinsic cellular defense mechanism, acts as a major obstacle to successful editing. MMR refers to the ability of cells to recognize and revert externally introduced variations as errors. When this mechanism is activated, gene editing efficiency can be significantly reduced. Researchers have continuously sought to identify nucleotide combinations that achieve the intended editing effect while evading the MMR surveillance. However, manually reviewing tens of thousands of possible scenarios has significant limitations in terms of time and cost.
Key Findings
The research team developed OptiPrime, a machine learning (ML) model that accurately predicts gene editing efficiency. This model applies the previously elucidated PE mechanisms to a computer algorithm. Based on experimental data learning, it predicts the efficiency of pegRNA with high accuracy. In particular, it includes the prediction of outcomes for single-target editing, as well as dual guide RNA (PE3) and twinPE methods.
OptiPrime can precisely analyze the nucleotide environment in which the mammalian MMR mechanism strongly reacts. Based on the learned information, the artificial intelligence can recommend silent edits that bypass the MMR system's surveillance. Silent edits refer to a technique that subtly alters the nucleotide sequence without changing the protein amino acid sequence, thereby cleverly evading the cell's repair mechanism. The researchers verified using primary human and mouse cells that applying the nucleotide combinations recommended by OptiPrime significantly increased gene editing efficiency in actual experiments.
Significance and Prospects
This research is expected to significantly reduce the random trial-and-error process required when designing gene editing therapeutics. Researchers can use a web server (https://optipri.me/) to pre-select the optimal nucleotide sequence candidates, greatly improving research efficiency. It is expected to shorten the design period from several months to a few days in the development of therapeutic agents for intractable genetic diseases that are targeted for clinical trials.
However, it should be noted that the predictive values of the computer model do not perfectly match all actual biological environments. Higher-order biological variables, such as cell type and chromatin accessibility, can still affect the prediction results. Furthermore, more validation data from diverse cell types is needed before it can be used in actual treatments for patients.
Prime editing (PE) can make specific local changes to genomic DNA in living systems but its efficient application currently requires extensive optimization of PE guide RNA (pegRNA) sequences. Here we present OptiPrime, a machine learning model of PE efficiency based on current understanding of PE mechanisms. OptiPrime achieves state-of-the-art accuracy on PE efficiency prediction and enables prediction of nicking guide RNA (PE3) and dual pegRNA (twinPE) outcomes. We validate that OptiPrime has learned the determinants of mammalian mismatch repair (MMR) and is well suited for nominating MMR-evasive silent edits that improve PE efficiency. We demonstrate the use of OptiPrime in a variety of prospective therapeutic contexts in primary human and mouse cells. Lastly, we show that OptiPrime can be used to achieve streamlined and efficient in vivo correction of a pathogenic mutation in the brain of a mouse model of KIF1A-associated neurological disorder. We provide a webserver for OptiPrime (https://optipri.me/) as a community resource.
This technology demonstrates its value in a scenario where it corrects a pathogenic mutation in the brain cells of a KIF1A-associated neurological disorder (KAND) mouse model. KAND is a rare and intractable genetic disease that causes central nervous system developmental disorders. Existing therapies have made it extremely difficult to completely deliver genes deep into brain nerve cells and successfully correct them with high efficiency within the complex brain tissue. The researchers successfully injected the optimal pegRNA selected by OptiPrime directly into the mouse brain, precisely correcting the base mutation that causes the disease. In the future, it is expected that this will lead to active clinical applications in the rapid acquisition of reliable optimal sequences for the design of precise gene therapies targeting human brain diseases or other solid organs.