Res2s2aM: Deep residual network-based model for identifying functional noncoding SNPs in trait-associated regions.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Noncoding single nucleotide polymorphisms (SNPs) contribute to disease heritability and polygenic traits.
• Identifying functional noncoding SNPs is crucial for understanding molecular mechanisms and advancing precision medicine.
• Genome-wide association studies (GWAS) identify trait-associated regions, but pinpointing causal noncoding SNPs remains challenging.

Purpose of the Study:

• To develop and evaluate a deep learning model for discriminating functional from nonfunctional noncoding SNPs.
• To leverage both DNA sequence context and SNP annotation data for improved SNP function prediction.
• To enhance the accuracy of identifying causal variants in post-GWAS analysis.

Main Methods:

• A deep residual network (ResNet)-based model, named Res2s2aM, was developed.
• The model integrates DNA sequence information with additional SNP annotation data.
• Performance was evaluated on a ground-truth set of disease-associated SNPs from the GRASP database.

Main Results:

• Res2s2aM significantly improves the prediction accuracy of functional noncoding SNPs.
• The model outperforms methods relying solely on sequence information.
• Res2s2aM shows superior performance compared to RegulomeDB, a leading tool for noncoding SNP prioritization.

Conclusions:

• The Res2s2aM model offers a powerful approach for identifying functional noncoding SNPs.
• Integrating sequence and annotation data enhances the prediction of SNP function.
• This work has implications for advancing precision medicine and understanding genetic contributions to disease.