RLASON-CDR: a reinforcement learning-driven adaptive synergistic optimization network for cancer drug response...

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Cancer drug response (CDR) prediction is vital for precision medicine.
• Existing computational methods struggle with semantic alignment of multimodal features, limiting CDR representation.
• Inconsistencies between multimodal and topological features reduce predictive accuracy.

Purpose of the Study:

• To develop a novel method for accurate and generalizable CDR prediction.
• To address limitations in semantic alignment and feature fusion in CDR prediction models.
• To enhance the biological interpretability of CDR prediction for clinical applications.

Main Methods:

• A Reinforcement Learning-driven Adaptive Synergistic Optimization Network-CDR (RLASON-CDR) was developed.
• RLASON-CDR constructs aligned multimodal CDR representations and captures topological features from cell line-drug networks.
• A reinforcement learning-based network adaptively optimizes these representations for synergistic pattern exploration.

Main Results:

• RLASON-CDR demonstrated superior performance over existing methods in CDR prediction.
• The model showed robustness in predicting responses for unknown drug-cell line combinations.
• Gradient attribution analysis identified key modality contributions and revealed biologically relevant response patterns.

Conclusions:

• RLASON-CDR effectively predicts cancer drug response by synergistically optimizing multimodal and topological features.
• The method offers biologically significant insights into drug response mechanisms.
• RLASON-CDR provides valuable guidance for advancing precision therapy in clinical settings.