Discovery of Dynamic Models for AML Disease Progression from Longitudinal Multi-Modal Clinical Data Using Explainable...

Area of Science:

• Oncology
• Computational Biology
• Medical Informatics

Background:

• Acute Myeloid Leukemia (AML) is a complex, aggressive cancer with high mortality.
• Effective prediction models require integrating longitudinal patient data.
• Understanding disease dynamics is crucial for research and clinical applications.

Purpose of the Study:

• To develop a robust methodology for discovering dynamic predictive models of AML progression.
• To elucidate the clinical, genetic, and treatment features influencing AML disease dynamics.
• To create an explainable machine learning algorithm for predicting AML progression.

Main Methods:

• Utilized a novel longitudinal multimodal clinical dataset of AML patients.
• Employed high-performance evolutionary computation for an explainable machine learning algorithm.
• Discovered mathematical models, including interactions, parameters, and nodes, predictive of AML progression.

Main Results:

• The methodology accurately estimated AML clinical dynamics, specifically blast percentages.
• Predictions were validated for both training and novel patient cohorts.
• Identified key clinical, genetic, and treatment features modulating disease progression.

Conclusions:

• The explainable machine learning approach successfully predicts AML progression using heterogeneous longitudinal data.
• This methodology demonstrates significant potential for modeling other acute disease progression dynamics.
• Provides a flexible framework for advancing clinical and translational research in oncology.