frDSM

frDSM predicts deleterious synonymous mutations in the human genome by integrating feature representation learning and ensemble classifiers to distinguish deleterious from benign variants.

Key Features:

• Logistic regression core: Uses logistic regression as the primary predictive model to classify deleterious versus benign synonymous mutations.
• Feature representation learning: Integrates functional scores derived from existing computational methods, evolutionary conservation metrics, splicing-related features, and sequence-based descriptors.
• Ensemble of XGBoost classifiers: Processes input descriptors with 76 XGBoost classifiers that generate predictive probability values which are concatenated into a 76-dimensional feature vector.
• Feature selection: Applies feature selection to remove redundant and irrelevant features, retaining 31 optimal features for final prediction.

Scientific Applications:

• Disease mechanism elucidation: Identifies deleterious synonymous mutations to support investigation of functional mechanisms underlying genetic diseases.
• Genomic research: Provides predictive annotations of synonymous variants to inform studies of gene function and regulation in the human genome.

Methodology:

Integrates functional scores from existing computational methods, evolutionary conservation metrics, splicing-related features, and sequence-based descriptors via a feature representation learning approach; processes descriptors with 76 XGBoost classifiers to produce predictive probability values concatenated into a 76-dimensional vector; employs logistic regression as the core classifier; and performs feature selection to retain 31 optimal features.