SHARE
Research Highlight

Exploring Protein Self-Organization with Machine Learning

Exploring Protein Self-Organization with Machine Learning
Schematic of the developed workflow.

Scientific Achievement

A machine learning workflow was developed for studying the dynamics of complex ordering systems with active rotational degrees of freedom.

Significance and Impact

The developed workflow is universally applicable for the description of optical, scanning probe, and electron microscopy imaging data and seeks to understand the dynamics of complex systems where rotations are a significant part of the process.

Research Details

- Dynamics of protein nanorods are visualized via high-speed atomic force microscopy (AFM). - Deep convolutional neural networks (DCNN) are used to convert noisy AFM observations into particle trajectories. - A rotationally invariant variational autoencoder (rVAE) is applied to the DCNN output and disentangles representations in the latent space revealing a rich spectrum of local transitions.   S. V. Kalinin, S. Zhang, M. Valleti, H. Pyles, D. Baker, J. J. De Yoreo, and M. Ziatdinov, “Disentangling Rotational Dynamics and Ordering Transitions in a system of Self-Organizing Protein Nanorods via Rotationally Invariant Latent Representations,” ACS Nano (2021). DOI: 10.1021/acsnano.0c08914