ACDA

ACDA predicts drug synergy by applying machine learning to molecular and pharmacological data to prioritize effective cancer drug combinations.

Key Features:

• Machine Learning Integration: Employs random forest regression to model drug–drug interactions and improve synergy prediction accuracy.
• Data Utilization: Integrates drug target information, gene mutations, and monotherapy drug sensitivity data from the Cancer Drug Atlas (CDA).
• Performance Enhancement: Augmentation with random forest regression and hyper-parameter tuning via cross-validation yields a reported 68% performance improvement when validated across datasets spanning ten tissues.
• Benchmarking Success: Demonstrates superior performance in comparative analyses, outperforming other methods in 16 out of 19 cases including benchmarks from the DREAM Drug Combination Prediction Challenge.
• Expanded Training Capabilities: Trained additionally on the Novartis Institutes for BioMedical Research PDX encyclopedia to extend prediction capability to Patient-Derived Xenograft (PDX) models.
• Visualization: Implements a novel visualization approach for interpreting synergy-prediction data.

Scientific Applications:

• Drug synergy prediction in cancer models: Predicts synergistic drug combinations for cancer cell-line and preclinical models.
• Combination therapy optimization: Prioritizes drug combinations to support optimization of therapies aimed at overcoming drug resistance and improving patient responses.
• PDX sensitivity prediction: Generates sensitivity predictions for Patient-Derived Xenograft (PDX) models to support preclinical research.
• Method benchmarking: Serves as a platform for comparative evaluation against established datasets and challenges, including the DREAM Drug Combination Prediction Challenge.

Methodology:

Integrates molecular (e.g., gene mutations) and pharmacological data (drug targets, monotherapy sensitivity); applies random forest regression; performs hyper-parameter tuning via cross-validation; and validates/benchmarks predictions against established datasets and competing methods.