MMSyn: A New Multimodal Deep Learning Framework for Enhanced Prediction of Synergistic Drug Combinations

Area of Science:

• Oncology
• Computational Biology
• Pharmacology

Background:

• Combination therapy offers a promising strategy for effective cancer treatment.
• Screening synergistic drug combinations in vitro is laborious and expensive due to the vast number of possibilities.
• Deep learning (DL) presents a viable computational approach for predicting synergistic drug combinations, leveraging high-throughput screening data.

Purpose of the Study:

• To propose and evaluate a multimodal deep learning framework, MMSyn, for predicting synergistic drug combinations.
• To demonstrate the efficacy of MMSyn in identifying potent drug combinations for cancer therapy.

Main Methods:

• MMSyn integrates drug molecular features (structure, fingerprint, string encoding) with cancer cell line data (gene expression, DNA copy number, pathway activity).
• Features are processed and integrated using an attention mechanism and an interaction module.
• A multilayer perceptron is employed for the final prediction of drug synergy.

Main Results:

• MMSyn significantly outperformed five state-of-the-art DL methods and three traditional machine learning models in drug combination prediction.
• The model demonstrated superior performance in stratified cross-validation across both drug combination and cell line datasets.
• Ablation experiments confirmed the effectiveness of individual components within the MMSyn framework.

Conclusions:

• MMSyn is a powerful and effective tool for predicting synergistic drug combinations.
• The multimodal approach enhances the accuracy and reliability of drug synergy prediction.
• MMSyn has the potential to accelerate the discovery of novel combination therapies for cancer treatment.