Machine Learning-Based Potentials for Digital Twins of Soft Matter

Abstract

Soft matter, encompassing colloidal suspensions, polymers, cells, and tissues, exhibits complex responses to external stimuli. Understanding and predicting this behavior is crucial for designing new materials and technologies. Digital twins, virtual models mimicking real systems, offer a powerful solution.
This work proposes a novel approach for developing accurate digital twins of soft matter systems by leveraging machine learning-based potentials. Our methodology involves: (i) extracting quantitative parameters from experimental data using deep learning algorithms; (ii) constructing interaction potentials between particles or agents using machine learning; (iii) building digital models that simulate system dynamics using these learned potentials; and (iv) validating model predictions against experimental results.
The core innovation lies in the development of machine learning-based potentials that accurately describe interactions within soft matter systems. This allows for virtual experimentation, optimization of self-assembly processes, prediction of material properties, and accelerated research and development in various fields.
We demonstrate the effectiveness of our approach by analyzing images of astrocytes, neurons, cell spheroids, and colloidal microparticles. This paves the way for developing digital twins of cell ensembles and colloidal microparticle systems, capable of simulating system behavior under diverse conditions. This offers valuable insights into soft matter dynamics and opens new possibilities for designing materials and technologies with tailored properties.

Funding: Russian Science Foundation (Grant No. 22-72-10128)

Speaker

Ivan V. Simkin
Bauman Moscow State Technical University
Russia

Discussion

Ask question