MutBLESS

MutBLESS predicts disease-prone sites within cancer genomes using deep neural networks applied to amino acid sequence-based features to characterize mutation impact.

Key Features:

• Data Collection: Utilizes a dataset of experimentally known driver mutations across 22 cancer types categorized into six groups (BRCA, LAML, EC, STAD, SKCM, and other cancer types), comprising 5,747 disease-prone sites and 5,514 neutral sites in 516 proteins.
• Motif Analysis: Analyzes amino acid distribution at mutant sites and identifies motif enrichment, with AAA and LR motifs prevalent in disease-prone sites and QPP and QF dominant in neutral sites.
• Deep Neural Network Methodology: Employs deep neural networks that use amino acid sequence-based features including physicochemical properties, secondary structure, tri-peptide motifs, and conservation scores to predict disease-prone sites.
• Performance Metrics: Reports an average AUC of 0.97 for BRCA, LAML, EC, STAD, and SKCM in a test dataset, an AUC of 0.72 across other cancer types combined, and sensitivity 96.56%, specificity 97.39%, and accuracy 97.64% for identifying cancer-specific mutations.

Scientific Applications:

• Identification of disease-prone sites: Identifies candidate disease-prone amino acid sites in cancer genomes for downstream experimental validation.
• Targeted therapy development: Informs development of targeted therapeutic strategies by highlighting mutation-prone motifs and sites.
• Mutation pattern analysis: Provides insights into motif enrichment and mutation patterns associated with oncogenesis across cancer types.
• Precision medicine support: Prioritizes mutations with high predicted disease relevance to support precision medicine efforts.

Methodology:

Uses an experimentally curated driver mutation dataset across 22 cancer types grouped into six categories; analyzes amino acid distribution and motif enrichment at mutant sites; extracts sequence-based features (physicochemical properties, secondary structure, tri-peptide motifs, conservation scores); and trains/predicts with deep neural networks.