TVAR: assessing tissue-specific functional effects of non-coding variants with deep learning.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Accurate functional annotation of non-coding variants from whole-genome sequencing (WGS) is crucial for understanding genetic disease risk.
• Rare non-coding variants are challenging to analyze but may play a significant role in disease predisposition.

Purpose of the Study:

• To develop a deep learning model, TVAR, for predicting the functionality of non-coding variants using epigenomic features.
• To leverage expression quantitative trait loci (eQTLs) across multiple human tissues to identify regulatory variants.

Main Methods:

• TVAR employs a multi-label learning deep neural network approach.
• It integrates thousands of tissue- and cell-type-specific epigenomic features with eQTL data from the GTEx project across 49 human tissues.
• The model accounts for correlations among tissues to capture shared and tissue-specific regulatory effects.

Main Results:

• TVAR achieves an average AUROC of 0.77 in predicting variant functionality across tissues.
• It demonstrates superior performance in identifying functional common and rare variants for complex diseases compared to state-of-the-art tools.
• TVAR shows consistent high performance across various validation datasets, including ClinVar, GWAS loci, and MPRA-validated variants.

Conclusions:

• TVAR effectively predicts the functional impact of non-coding variants by integrating multi-tissue epigenomic data.
• The tool enhances the analysis of WGS data for identifying genetic variants associated with complex human diseases.
• TVAR provides valuable tissue-specific functional annotations for rare variants, advancing genetic research.