From data to QSP models: a pipeline for using Boolean networks for hypothesis inference and dynamic model building.

Area of Science:

• Pharmacology and Systems Biology
• Computational Biology and Bioinformatics

Background:

• Quantitative Systems Pharmacology (QSP) models traditionally rely on hypothesis-driven, knowledge-based approaches to define network structures.
• Advances in generating large, hypothesis-neutral datasets present an opportunity for data-driven model development.
• Existing QSP model development is often limited by the need for extensive prior knowledge.

Purpose of the Study:

• To explore a data-driven pipeline for constructing complex network representations of physiological responses to pharmaceuticals.
• To convert logic-based data analysis into dynamic ordinary differential equation (ODE)-based Quantitative Systems Pharmacology (QSP) models.
• To demonstrate a robust and repeatable method for inferring biological network topologies from data without prior system knowledge.

Main Methods:

• An integrated pipeline was developed, starting with k-means clustering to binarize continuous data.
• A Best-Fit Extension method was employed to infer network relationships and construct a Boolean network.
• The Boolean network was subsequently converted into a continuous ODE-based QSP model.

Main Results:

• The pipeline successfully reconstructed a QSP model using data generated from an existing flotetuzumab QSP model.
• The reconstructed model exhibited no false-positive relationships.
• The output dynamics of key species in the reconstructed model closely matched the original QSP model.

Conclusions:

• The developed pipeline enables accurate inference of biological system relationships in a hypothesis-neutral manner.
• This approach facilitates the creation of robust QSP models from large datasets without requiring prior system knowledge.
• The method demonstrates significant potential for advancing drug response modeling and discovery.