Mechanistically Explainable AI Model for Predicting Synergistic Cancer Therapy Combinations.

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Predicting synergistic drug combinations is crucial for effective cancer therapy.
• Current methods often lack mechanistic insights and struggle with large datasets.
• Integrating diverse data sources can enhance predictive accuracy.

Purpose of the Study:

• To develop and validate a Large Language Model (LLM)-based framework for predicting synergistic oncology drug combinations.
• To provide mechanistic insights into drug synergy.
• To improve the efficiency of cancer drug discovery and translational strategies.

Main Methods:

• Developed a retrieval-augmented generation (RAG) framework utilizing Large Language Models (LLMs).
• Integrated over 50,000 in vitro drug pair assay results and 1631 clinical/preclinical trial entries.
• Combined drug combination data with a knowledge graph for enhanced predictions.

Main Results:

• The framework achieved a high predictive accuracy, demonstrated by an F1 score of 0.80.
• Successfully integrated extensive in vitro and clinical data for improved model performance.
• Provided mechanistic explanations for predicted drug synergies.

Conclusions:

• The LLM-based RAG framework shows significant potential for streamlining oncology drug discovery.
• This approach enhances predictive accuracy and explainability in identifying synergistic drug combinations.
• The study highlights a promising strategy for improving cancer treatment translational research.