Cosynllm: predicting drug combination synergy with LLM-generated descriptions

Area of Science:

• Computational Biology
• Pharmacology
• Artificial Intelligence

Background:

• Drug combination therapy is crucial for complex diseases but faces challenges due to the vast number of potential combinations.
• Experimental screening of all drug combinations is impractical and expensive.
• Deep learning and Large Language Models (LLMs) show potential for predicting synergistic drug combinations.

Purpose of the Study:

• To develop an LLM-assisted framework, CoSynLLM, for predicting drug combination synergy.
• To leverage LLM-derived semantic information and drug fingerprints for comprehensive drug representation.
• To integrate cell line gene expression data for cellular context in synergy prediction.

Main Methods:

• CoSynLLM utilizes LLMs to generate semantic chemical information.
• Drug fingerprints are incorporated for explicit structural features.
• A hierarchical feature fusion strategy merges drug and cell line data for synergy prediction.

Main Results:

• CoSynLLM demonstrated competitive performance on benchmark datasets (NCI-ALMANAC and O'Neil).
• The framework effectively predicts drug combination synergy.
• The study highlights the utility of LLMs in computational drug discovery.

Conclusions:

• CoSynLLM provides a robust computational framework for predicting synergistic drug combinations.
• This approach can accelerate the identification of effective combination therapies.
• LLM-assisted methods offer a practical solution to the challenge of combinatorial drug screening.