MERIDA

MERIDA predicts drug sensitivity in cancer cells by learning interpretable Boolean logic rules from multi-omics (genetic and molecular) data using a modified integer linear programming formulation.

Key Features:

• Interpretable Models: Generates Boolean logic-based models to represent rules that link molecular features to drug sensitivity or resistance.
• Enhanced Computational Efficiency: Alters logical rules from the LOBICO method to achieve accelerated running times and enable larger input feature sets.
• Comprehensive Modeling: Supports integration of diverse multi-omics data to model mechanisms related to cancer initiation, progression, and drug response.
• Incorporation of A Priori Knowledge: Enables inclusion of existing biomarker databases or iteratively acquired knowledge from previous runs to inform model construction.
• Identification of Biomarkers: Identifies potential sensitivity and resistance biomarkers from the learned logical rules.

Scientific Applications:

• Personalized oncology: Tailors drug-sensitivity predictions to genetic and molecular profiles of cancer samples for individualized treatment stratification.
• Biomarker discovery: Discovers candidate sensitivity and resistance biomarkers from multi-omics-derived logical rules.
• Modeling cancer mechanisms: Models interactions among molecular features to study processes related to cancer initiation and progression.
• Method benchmarking: Provides a framework for comparing predictive performance against other machine learning or rule-based methods.

Methodology:

Uses a modified integer linear programming formulation built on and altering LOBICO, constructs Boolean logic-based rule models, and supports iterative refinement through successive runs and incorporation of prior biomarker knowledge.