DeepMutSim: exhaustive in silico enumeration of human gene SNVs and amino-acid substitutions based on MANE Select transcripts in HGVS nomenclature
Abstract
Background The accurate annotation and interpretation of genetic variants are fundamental to genomic medicine. However, researchers who design functional assays, deep mutational scanning experiments, or machine-learning benchmarks for a gene of interest must first obtain an exhaustive list of the theoretically possible variants of that gene, a preparatory step for which dedicated tools are lacking. DeepMutSim performs exhaustive in silico enumeration of predefined variant classes; it does not detect variants from sequencing data and does not provide functional, pathogenicity, population-frequency, or clinical interpretation. Methods We developed DeepMutSim, an open-source Python package for the exhaustive in silico enumeration of single-nucleotide variants (SNVs) and single amino-acid substitutions. Given a human gene symbol, DeepMutSim retrieves the MANE Select transcript (optionally MANE Plus Clinical) as a canonical reference and enumerates all three alternative bases at every nucleotide of the amino-acid-encoding portion of the coding sequence (CDS; the terminal stop codon is excluded by design), the 5’ and 3’ untranslated regions (UTRs), the canonical splice-site dinucleotides, and the extended intronic splice-site regions, together with all 19 amino-acid substitutions per residue and all 63 alternative codons per codon. All variants are described using the official Human Genome Variation Society (HGVS) nomenclature and are accompanied by their genomic coordinates (GRCh38). Reference bases at intronic splice-site positions are read from the RefSeqGene (NG_) record rather than assumed. Results Applied to the G6PD MANE Select transcript (NM_001360016.2), DeepMutSim enumerated 4,635 CDS SNVs, 198 5’UTR SNVs, 1,827 3’UTR SNVs, 132 canonical splice-site SNVs, 1,122 extended splice-site-region SNVs, 9,785 amino-acid substitutions, and 32,445 codon substitutions (29,668 of which are missense). The generated variant set was validated with an 80-test offline unit suite and by checking approximately 400,000 genomic coordinates of generated variants against the reference chromosome sequence across five genes (INS, TMPRSS6, ADAMTS13, SLC22A5, CDKN2A) spanning both strands, with zero mismatches. The lightweight, dependency-minimized architecture enables integration into existing bioinformatics pipelines for deep mutational scanning (DMS) design, machine-learning training, and the characterization of variant landscapes in disease-causing genes. Conclusions DeepMutSim addresses a gap in the genomics community by providing a standardized and reproducible tool for exhaustive variant enumeration. By anchoring its output to the MANE standard, HGVS nomenclature, and GRCh38 coordinates, DeepMutSim produces computationally interoperable variant sets that support a wide range of downstream research applications.