SKM 2021 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

BP: Fachverband Biologische Physik

BP 9: Machine Learning in Dynamical Systems and Statistical Physics (joint session DY/BP)

BP 9.2: Talk

Friday, October 1, 2021, 11:30–11:45, H2

Employing artificial neural networks to find reaction coordinates and pathways for self-assembly — •Jörn Appeldorn, Arash Nikoubashman, and Thomas Speck — Inst. für Physik, Universität Mainz, Germany

We study the spontaneous self-assembly of single-stranded DNA fragments using the coarse-grained oxDNA2 implementation [1]. A successful assembly is a rare event that requires crossing a free energy barrier. Advanced sampling methods like Markov state modeling allow to bridge these long time scales, but they require one or more collective variables (order parameters) that faithfully describe the transition towards the assembled state. Formulating an order parameter typically relies on physical insight, which is then verified, e.g., through a committor analysis. Here we explore the use of autoencoder neural networks to automatize this process and to find suitable collective variables based on structural information. For this step, one still needs to map configurations onto structural descriptors, which is a non-trivial task. Specifically, we investigate the latent space of EncoderMap [2] and how it changes with the amount of information contained in the descriptor. With this approach, we were able to determine the free energy landscape, the locations of the (meta)stable states, and the corresponding transition probabilities.

[1] - Snodin et al., J. Chem. Phys.(2015), 142, 234901 [2] - T. Lemke and and C. Peter,J.Chem.TheoryComput.(2019), 15, 1209-1215