DrugnomeAI

DrugnomeAI predicts the druggability likelihood of protein-coding genes across the human exome using an ensemble machine-learning framework for drug target selection.

Key Features:

• Ensemble stochastic semi-supervised learning: Applies an ensemble stochastic semi-supervised learning framework to predict gene druggability exome-wide.
• Feature integration from 15 sources (324 features): Integrates data from 15 distinct sources to generate 324 gene-level features characterizing each protein-coding gene.
• Exome-wide coverage: Produces druggability scores for every protein-coding gene in the human exome.
• High predictive accuracy: Achieves a median area under the curve (AUC) of 0.97 for exome-wide predictions.
• Protein–protein interaction predictors: Features derived from protein-protein interaction networks are identified as top predictors of druggability.
• Generic and specialized models: Provides both generic models and specialized models tailored to specific disease types or therapeutic modalities.
• Validation against clinical and phenotypic data: Top-ranking genes are significantly enriched in clinical development programs (p < 1 × 10^-308) and achieve genome-wide significance in phenome-wide association studies of UK Biobank exomes for binary (p = 1.7 × 10^-5) and quantitative traits (p = 1.6 × 10^-7).

Scientific Applications:

• Drug target prioritization: Prioritizes candidate therapeutic targets by providing druggability likelihoods across the human exome.
• Clinical candidate selection support: Identifies genes enriched in clinical development programs to inform target selection decisions.
• Interpretation of exome-wide association results: Supports interpretation of phenome-wide association study findings from UK Biobank exomes by linking genetic associations to druggability.

Methodology:

Implements an ensemble stochastic semi-supervised learning framework trained on 324 features derived from 15 data sources, uses protein–protein interaction network features among predictors, and validates predictions via enrichment analysis against clinical development programs and phenome-wide association studies on UK Biobank exomes.