CDFA: Calibrated deep feature aggregation for screening synergistic drug combinations

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence

Background:

• Drug combination therapy offers improved efficacy and safety for complex diseases.
• Identifying optimal drug combinations is challenging due to the vast search space, necessitating computational approaches.
• Machine learning and deep learning are emerging as powerful tools for navigating drug combination screening.

Purpose of the Study:

• To introduce a novel computational framework, Calibrated Deep Feature Aggregation (CDFA), for screening synergistic drug combinations.
• To develop a robust and reliable method for predicting effective drug pairs in combination therapy.
• To address the limitations of conventional wet-lab experimentation in drug discovery.

Main Methods:

• CDFA utilizes a novel cell line representation integrating protein and gene expression data.
• A Transformer-based feature aggregation network with multi-head attention models drug-pair and cell-line interactions.
• Uncertainty quantification and calibration are incorporated to enhance prediction reliability.

Main Results:

• CDFA demonstrates superior performance compared to existing state-of-the-art deep learning models.
• Experimental results validate the effectiveness of the proposed framework in identifying synergistic drug combinations.
• The model successfully captures complex non-linear drug-cell interactions.

Conclusions:

• CDFA provides a computationally efficient and reliable tool for drug combination screening.
• The biologically informed cell line representation and attention mechanisms contribute to CDFA's superior performance.
• Uncertainty calibration enhances the trustworthiness of predicted synergistic drug combinations, aiding drug discovery efforts.