DSA-DeepFM: a dual-stage attention-enhanced DeepFM model for predicting anticancer synergistic drug combinations

Area of Science:

• Computational biology
• Pharmacology
• Machine learning

Background:

• Drug combinations are essential for overcoming drug resistance and enhancing therapeutic outcomes.
• The vast number of potential drug combinations necessitates computational approaches beyond traditional experimental methods.
• Existing machine learning models often overlook biological mechanisms, limiting their predictive accuracy for synergistic drug combinations.

Purpose of the Study:

• To develop a deep learning model, DSA-DeepFM, that accurately predicts synergistic two-drug combinations by integrating biological feature interactions.
• To improve the prediction of synergistic drug combinations by incorporating both categorical and numerical data.
• To address the limitations of current methods by considering complex biological mechanisms.

Main Methods:

• Developed DSA-DeepFM, a deep learning model combining a dual-stage attention (DSA) mechanism with Factorization Machines (FMs).
• Integrated categorical and auxiliary numerical inputs to capture field-aware and embedding-aware patterns.
• Utilized a deep FM module to process feature interactions for prediction.

Main Results:

• DSA-DeepFM significantly outperformed traditional machine learning and state-of-the-art deep learning models in predicting synergistic drug combinations.
• t-SNE visualizations confirmed the model's discriminative power.
• Identified eight novel synergistic drug combinations, demonstrating practical utility.

Conclusions:

• DSA-DeepFM offers a powerful and accurate approach for predicting synergistic drug combinations.
• The model's ability to integrate complex biological interactions enhances its predictive performance.
• The identified novel synergistic combinations highlight the model's potential for drug discovery and development.