MtBNN

MtBNN predicts the functional impact of non-coding single-nucleotide variants by using a multi-task Bayesian neural network to integrate multi-omic chromatin-profiling data and quantify regulatory disruption.

Key Features:

• Multi-task Bayesian Framework: Employs a multi-task Bayesian neural network framework to integrate diverse omic and chromatin-profiling data and extract features such as chromatin accessibility and transcription factor binding affinities.
• Quantification of Deleterious Impact: Calculates probability distributions indicating whether a non-coding variant disrupts regulatory activities and quantifies functional impact across contexts including expression quantitative trait loci (eQTL), DNase I sensitivity quantitative trait loci (DNaseI-QTL), and ATAC-seq peaks.
• Performance Superiority: Demonstrates superior performance versus existing methods in identifying functional non-coding variants and in fine-mapping genome-wide association study (GWAS) SNPs to prioritize potentially causal disease-associated variants.

Scientific Applications:

• Gene Regulation Studies: Provides insights into how variation in non-coding regulatory regions and regulatory element function influence gene regulation.
• Disease Association Mapping: Aids fine-mapping of GWAS SNPs to identify disease-associated non-coding variants.
• Regulatory Variant Prioritization: Prioritizes regulatory variants with substantial impacts on human phenotypes for downstream genetic studies.

Methodology:

Applies a Bayesian deep learning multi-task neural network to integrate multi-omic chromatin-profiling data, extract features from genomic sequences, and compute probability distributions of non-coding variant impacts.