Using predictive machine learning models for drug response simulation by calibrating patient-specific pathway...

Area of Science:

• Computational Biology and Bioinformatics
• Machine Learning in Medicine
• Pharmacogenomics

Background:

• Pathway signatures are crucial for identifying dysregulated biological processes in patients.
• Machine learning (ML) methods utilize pathway signatures for applications like precision medicine, drug repurposing, and drug discovery.
• Predicting individual patient drug response remains a significant challenge in clinical practice.

Purpose of the Study:

• To develop and validate a novel ML-based methodology for simulating drug responses in individual patients.
• To utilize pathway activity scores and an intuitive scoring algorithm to predict treatment outcomes.
• To establish a computational proxy for identifying potential drug candidates and understanding drug mechanisms.

Main Methods:

• Leveraged highly predictive ML models calibrated with pathway activity scores from disease samples.
• Employed an intuitive scoring algorithm to modify patient pathway signatures, simulating drug treatment effects.
• Evaluated the model's ability to predict a shift from diseased to normal state post-simulated drug treatment.

Main Results:

• Successfully identified approved and investigational drugs for four different cancer types.
• Outperformed six comparable state-of-the-art methods in drug identification accuracy.
• Demonstrated the capability to deconvolute drug mechanisms of action and propose effective combination therapies.

Conclusions:

• The developed methodology shows significant promise for supporting clinical decision-making in personalized medicine.
• Simulating drug effects on individual patients via pathway signature modification offers a powerful tool for precision oncology.
• This approach advances drug discovery, repurposing, and the understanding of therapeutic interventions.