A Graph-Based Machine-Learning Approach Combined with Optical Measurements to Understand Beating Dynamics of...

Area of Science:

• Computational biology
• Biophysics
• Machine learning

Background:

• Predicting cardiac cellular dynamics computationally is difficult due to the absence of first-principle mathematical models.
• Existing models often struggle with accuracy and scalability for heterogeneous cell populations.

Purpose of the Study:

• To develop a novel machine learning approach for robust prediction of cardiomyocyte dynamics.
• To integrate physics simulation with graph networks for enhanced predictive power.
• To validate the model using in vitro experimental data.

Main Methods:

• Developed a constraint-based interaction neural projection (CINP) algorithm, a hybrid physics-simulation and graph network approach.
• Embedded inductive physical priors into the model to learn from sparse image data.
• Utilized an in vitro platform for cellular motion and calcium transient analysis to validate predictions.

Main Results:

• The CINP algorithm successfully uncovered hidden physical constraints from limited cardiac cell data.
• The model provided robust predictions for heterogeneous, large-scale cell sets.
• Validated model efficacy in predicting complex organoid cellular behaviors in both silico and in vitro settings.

Conclusions:

• The proposed hybrid machine learning model offers a powerful new tool for predicting cardiac cellular dynamics.
• This approach overcomes limitations of traditional models by integrating physical principles and learning from data.
• The validated model demonstrates potential for advancing research in cardiac cell biology and disease modeling.