MVP predicts the pathogenicity of missense variants by deep learning

Abstract

Accurate pathogenicity prediction of missense variants is critically important in genetic studies and clinical diagnosis. Previously published prediction methods have facilitated the interpretation of missense variants but have limited performance. Here, we describe MVP (Missense Variant Pathogenicity prediction), a new prediction method that uses deep residual network to leverage large training data sets and many correlated predictors. We train the model separately in genes that are intolerant of loss of function variants and the ones that are tolerant in order to take account of potentially different genetic effect size and mode of action. We compile cancer mutation hotspots and de novo variants from developmental disorders for benchmarking. Overall, MVP achieves better performance in prioritizing pathogenic missense variants than previous methods, especially in genes tolerant of loss of function variants. Finally, using MVP, we estimate that de novo coding variants contribute to 7.8% of isolated congenital heart disease, nearly doubling previous estimates.

Predicting functional consequences of mutations using molecular interaction network features

“…Performance was compared to 24 state-of-the-art methods that do not use network-based information, 4 cancer-focused methods: CHASM (Carter et al 2009), ParsSNP (Kumar et al 2016), TransFIC (Gonzalez-Perez et al 2012, and CanDrA (Mao et al 2013), and 20 population-based methods: VEST (Carter et al 2013), SIFT (Ng and Henikoff 2003), PolyPhen (Adzhubei et al 2010), CADD (Kircher et al 2014), ClinPred (Alirezaie et al 2018), DANN (Quang et al 2015), DEOGEN2 (Raimondi et al 2017), FATHMM (inherited disease version) (Shihab et al 2013), LIST-S2 (Malhis et al 2020), LRT (Chun and Fay 2009), M-CAP (Jagadeesh et al 2016), MPC (Samocha et al 2017), MVP (Qi et al 2021), Met-aLR and MetaSVM (Dong et al 2015), MutPred (Pejaver et al 2020), MutationAssessor (Reva et al 2011), Muta-tionTaster (Schwarz et al 2014), PROVEAN (Choi et al 2012), and REVEL (Ioannidis et al 2016). We obtained prediction scores for the mutations in the (Tokheim and Karchin 2019).…”

Artificial intelligence enables comprehensive genome interpretation and nomination of candidate diagnoses for rare genetic diseases

“…Exomiser [ 16 ] benchmark analyses were run with the same configuration used in the 100,000 Genomes Project [ 77 ], specifically (1) using the GRCh37 genome assembly; (2) analyzing autosomal and X-linked forms of dominant and recessive inheritance; (3) allele frequency sources from the 1000 Genomes Project [ 78 ], TopMed [ 79 ], UK10K [ 80 ], ESP, ExAC [ 81 ], and gnomAD [ 73 ] (except Ashkenazi Jewish); (4) pathogenicity sources from REVEL [ 82 ] and MVP [ 83 ]; and (5) including the steps failedVariantFilter, variantEffectFilter (remove non-coding variants), frequencyFilter with maxFrequency = 2.0, pathogenicityFilter with keepNonPathogenic = true, inheritanceFilter, omimPrioritiser, and hiPhivePrioritiser.…”

Predicting functional effect of missense variants using graph attention neural networks

“…Numerous methods, such as Polyphen 9 , SIFT 10 , CADD 11 , REVEL 12 , MetaSVM 13 , M-CAP 14 , Eigen 15 , MVP 16 , PrimateAI 17 , MPC 18 , and CCRs 19 , have been developed to address the problem. These methods differ in several aspects, including the prediction features, how the features are represented in the model, the training data sets, and how the model is trained.…”