Using machine learning surrogate modeling for faster QSP VP cohort generation.

Area of Science:

• Pharmacology
• Computational Biology
• Machine Learning

Background:

• Virtual patients (VPs) are crucial in quantitative systems pharmacology (QSP) for assessing variability and uncertainty in clinical responses.
• Traditional VP generation methods, involving random parameter sampling and output-based rejection, are often inefficient, with most runs failing to produce valid VPs.

Purpose of the Study:

• To introduce and demonstrate a novel workflow for significantly enhancing the efficiency of virtual patient (VP) creation in quantitative systems pharmacology (QSP) modeling.
• To showcase the application of machine learning surrogate models for rapid pre-screening of parameter combinations.
• To provide a tutorial on selecting and optimizing surrogate models using a dedicated software application.

Main Methods:

• Training machine learning surrogate models using the full QSP model.
• Employing surrogate models to pre-screen parameter combinations for feasibility.
• Validating pre-screened parameter combinations with the original QSP model.
• Utilizing a software application for surrogate model selection and optimization.

Main Results:

• The surrogate model approach dramatically improves the efficiency of VP generation compared to traditional methods.
• The vast majority of parameter combinations pre-vetted by surrogate models result in valid VPs when tested in the original QSP model.
• The tutorial demonstrates a practical workflow for implementing this method.

Conclusions:

• Machine learning surrogate models represent a significant advancement in the efficient generation of virtual patients for QSP.
• The proposed workflow offers a scalable and effective solution for exploring parameter spaces in complex biological models.
• This approach facilitates more robust investigations into drug response variability and uncertainty.