TVAR

TVAR predicts tissue-specific functional effects of non-coding genetic variants using a multi-label deep learning model trained on epigenomic features and eQTLs across 49 human tissues.

Key Features:

• Multi-label deep neural network: Employs a multi-label learning-based deep neural network architecture that integrates high-dimensional epigenomic features with eQTL data from 49 human tissues (GTEx).
• Tissue-specific and shared effects: Captures both shared and tissue-specific regulatory roles of non-coding variants by modeling correlations among different tissues.
• Performance metrics: Achieves an average area under the receiver operating characteristic curve (AUROC) of 0.77 across multiple tissues.
• Disease-specific evaluation: Evaluated on coronary artery disease, breast cancer, Type 2 diabetes, and schizophrenia, demonstrating superior performance over five existing state-of-the-art tools for common and rare variants.
• G-score aggregation: Implements a G-score that aggregates tissue-specific annotations across all tissues to provide comprehensive variant functional assessments.
• Comparative validation: Validated against datasets including ClinVar, fine-mapped GWAS loci, and MPRA-validated variants, with consistent outperformance of competing tools.

Scientific Applications:

• Complex disease genetics: Provides tissue-specific functional annotations to investigate how non-coding variants contribute to disease risk and progression in complex diseases.
• Variant prioritization in WGS studies: Enables prioritization of non-coding variants from whole-genome sequencing by predicting tissue-specific regulatory functionality.
• Regulatory mechanism elucidation: Integrates multi-tissue epigenomic data and eQTLs to elucidate tissue-dependent regulatory roles of genetic variants.

Methodology:

Uses a multi-label learning-based deep neural network trained on eQTLs from GTEx across 49 human tissues, integrating high-dimensional tissue-specific and shared epigenomic features.