MetaRNN: Differentiating Rare Pathogenic and Rare Benign Missense SNVs and InDels Using Deep Learning

With advances in high-throughput DNA sequencing, numerous genetic variants have been discovered in the human genome. One challenge we face is interpreting these variants to help in disease screening, diagnosis, and treatment. While multiple computational approaches have been proposed to improve our understanding of genetic variants, their ability to identify rare pathogenic variants from rare benign ones is still lacking. Using context annotations and deep learning methods, we present pathogenicity prediction models, MetaRNN and MetaRNN-indel, to help identify and prioritize rare non-synonymous single nucleotide variants (nsSNVs) and non-frameshift insertion/deletions (nfINDELs). A recurrent neural network incorporating a +/- 1 codon window around the affected codon was combined with 28 high-level annotation scores and allele frequency features to develop the two proposed models. We use independent test datasets to demonstrate that these new models outperform state-of-the-art competitors and achieve a more interpretable score distribution. Importantly, prediction scores from the nsSNV-based and the nfINDEL-based models are comparable, enabling easy adoption of integrated genotype-phenotype association analysis methods. In addition, we provide pre-computed MetaRNN scores for all possible human nsSNVs and a Linux executable file for a fast one-stop annotation of nsSNVs and nfINDELs. All the resources are available at http://www.liulab.science/MetaRNN.