Data-Driven Discovery of Mathematical and Physical Relations in Oncology Data Using Human-Understandable Machine...

Area of Science:

• Oncology
• Computational Biology
• Machine Learning

Background:

• Mathematical models and differential equations have long been used to understand cancer processes.
• Advancements in cancer genome sequencing provide abundant data for new modeling approaches.
• Integrating biological data with physical models offers predictive power for cancer progression.

Purpose of the Study:

• To introduce a novel machine learning system for discovering mathematical and physical relationships in oncology data.
• To enhance the mechanistic understanding of cancer dynamics using heterogeneous clinical data.
• To develop a tool for improved clinical decision-making in oncology.

Main Methods:

• Utilized a machine learning system employing neural networks with mechanisms of competition, cooperation, and adaptation.
• The system simultaneously learns statistics and governing relations from multiple clinical data covariates.
• Focused on extracting human-understandable properties and features from complex datasets.

Main Results:

• The system identified nonlinear conservation laws in cancer kinetics and growth curves.
• Discovered symmetries in tumor phenotypic staging transitions and preoperative spatial distribution.
• Modeled nonlinear intracellular and extracellular pharmacokinetics of neoadjuvant therapies.

Conclusions:

• The machine learning system effectively extracts human-understandable representations of cancer dynamics.
• Demonstrated the system's capability to support clinical decision-making through data-driven insights.
• The tool serves as a bridge between data, modelers, data scientists, and clinicians in mathematical and computational oncology.