3Cnet: pathogenicity prediction of human variants using multitask learning with evolutionary constraints

Abstract

Motivation Improvements in next-generation sequencing have enabled genome-based diagnosis for patients with genetic diseases. However, accurate interpretation of human variants requires knowledge from a number of clinical cases. Additionally, manual analysis of each variant detected in a patient's genome requires enormous time and effort. To reduce the cost of diagnosis, various computational tools have been developed to predict the pathogenicity of human variants, but the shortage and bias of available clinical data can lead to overfitting of algorithms. Results We developed a pathogenicity predictor, 3Cnet, that uses recurrent neural networks to analyse the amino acid context of human variants. As 3Cnet is trained on simulated variants reflecting evolutionary conservation and clinical data, it can find disease-causing variants in patient genomes with 2.2 times greater sensitivity than currently available tools, more effectively discovering pathogenic variants and thereby improving diagnosis rates. Availability Codes (https://github.com/KyoungYeulLee/3Cnet/) and data (https://zenodo.org/record/4716879#.YIO-xqkzZH1) are freely available to non-commercial users. Supplementary information Supplementary data are available at Bioinformatics online.

How Functional Genomics Can Keep Pace With VUS Identification

“…ClinPred integrates normal population variant frequency data from genomAD with cleaner ClinVar annotations into existing pathogenicity scores, which performed better than other top tools across several metrics ( 131 ). 3Cnet, uses neural networks trained using population data (genomAD), conservation data (UniREF) and clinical data (ClinVar) to outperform most popular tools including REVEL, SIFT, and PolyPhen previously discussed ( 132 ).…”

Cutting-Edge AI Technologies Meet Precision Medicine to Improve Cancer Care

“…Sequence-based techniques [ 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 ] are used to construct the most common tools for predicting the pathogenicity of genetic variations. For instance, REVEL [ 10 ] employs a random forest method based on ensemble methods with 13 pathogenicity predictors.…”

“…Splice AI [ 13 ] can accurately predict splice junctions from an arbitrary pre-mRNA transcript sequence based on a deep neural network (DNN). Other novel prediction tools based on the utilized recurrent neural network, XGBoost (a variant of the gradient boosted tree), include 3Cnet [ 14 ] and VARITY [ 17 ]. The neural network can manage large training datasets and many correlated predictors.…”

“…Clinvitae et al [ 28 ] used penalized logistic regression to prioritize and classify the pathogenicity of genetic variants. We also created an ACMG/AMP-based score computation and designed unique algorithms to assign a score or degree of pathogenicity [ 14 ].…”

X-CAP improves pathogenicity prediction of stopgain variants

“…Despite the continuous accumulation of known pathogenic and benign variants in databases such as ClinVar [ 2 ] and the Human Gene Mutation Database (HGMD) [ 3 ], they are far from complete. For example, ClinVar has high-confidence pathogenicity labels for fewer than 100 thousand of all possible 82 million missense variants [ 4 ], and the HGMD collection grows by thousands of pathogenic variants every year [ 3 , 5 ]. This necessitates the development of computational tools that can distinguish pathogenic variants from benign ones.…”