MMFSyn: A Multimodal Deep Learning Model for Predicting Anticancer Synergistic Drug Combination Effect

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Combination therapy enhances efficacy and reduces toxicity.
• Identifying synergistic drug combinations is complex and challenging.
• Current methods struggle with the increasing number of drug combinations.

Purpose of the Study:

• To develop a novel deep learning model, MMFSyn, for predicting synergistic anti-cancer drug combinations.
• To integrate multimodal drug data with cancer cell line features for improved prediction accuracy.
• To address the challenge of identifying synergistic drug relationships in complex treatment regimens.

Main Methods:

• Utilized multimodal drug data (Morgan fingerprints, atom sequences, molecular diagrams, atomic point cloud) extracted via SMILES.
• Applied Bi-LSTM, gMLP, multi-head attention, and multi-scale GCNs for drug feature extraction.
• Integrated gene expression and mutation omics data from cancer cell lines to construct cell line features.
• Combined extracted features to predict synergistic anti-cancer drug combination effects.

Main Results:

• MMFSyn demonstrated superior performance compared to existing methods.
• Achieved a root mean square error (RMSE) of 13.33.
• Obtained a Pearson correlation coefficient (PCC) of 0.81, indicating strong predictive accuracy.

Conclusions:

• MMFSyn effectively captures complex relationships between multimodal drug data and omics features.
• The model significantly improves the prediction of synergistic anti-cancer drug combinations.
• Highlights the potential of deep learning in advancing drug discovery and combination therapy.