Regensburg 2022 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
BP: Fachverband Biologische Physik
BP 2: Computational Biophysics and Neuroscience
BP 2.3: Talk
Monday, September 5, 2022, 10:00–10:15, H13
RNA structure prediction via Machine Learning — •Alexander Schug1,2, Oskar Taubert4, Christian Faber1, Mehari Zerihun1, Fabrizio Pucci1, Fabrice von der Lehr3, Philipp Knechtges3, Marie Weiel4,5, Charlotte Debus4,5, Daniel Coquelin4,5, Stefan Kesselheim1,5, Achim Basermann3, Achim Streit4, and Markus Götz4,5 — 1Jülich Supercomputing Centre, FZ Jülich, Jülich — 2Faculty of Biology, University of Duisburg/Essen — 3Institute for Software Technology, German Aerospace Centre (DLR) — 4Steinbuch Centre for Computing, Karlsruhe Institute of Technology — 5Helmholtz AI
Knowledge of biomolecular structure is necessary to gain any detailed understanding of their function For proteins, tools rooted in statistical physics such as Direct Coupling Analysis (DCA) or Machine Learning driven approaches (ML) such as Alpha Fold 2 exploit massive sequence databases to trace evolutionary patterns for structure predictions. We demonstrate how additional information, such as low-resolution experimental information (e.g. SAXS or FRET) can integrated. For RNA there are significantly less data available than for proteins, which makes ML more challenging. Still, we demonstrate how contact prediction for RNA can be vastly improved both via simple convolutional neural networks but also by unsupervised deep-learning approaches by combining multiple self-supervised learning tasks. In an empirical evaluation for RNA, we find a strong increase of prediction quality.