EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
2726. Quality assessment of 3D RNA structures using graph neural networks
Invited abstract in session TC-20: Integrative Approaches in Health and Disease: From Molecular Structures to Clinical Outcomes, stream Computational Biology, Bioinformatics and Medicine.
Tuesday, 12:30-14:00Room: 45 (building: 116)
Authors (first author is the speaker)
| 1. | Maciej Antczak
|
| Institute of Computing Science, Poznan University of Technology | |
| 2. | Bartosz Adamczyk
|
| Institute of Computing Science, Poznan University of Technology | |
| 3. | Marta Szachniuk
|
| Institute of Computing Science, Poznan University of Technology |
Abstract
General quality assessment of 3D RNA structures predicted in silico is crucial to identify native or near-native 3D RNA folds. Nowadays, there are many, freely available computational methods for 3D RNA structure prediction, however without the reference - experimentally determined 3D RNA structure - it is difficult to reliably rank them in terms of practical usefulness. Here, we propose to apply graph neural networks to precisely infer the interatomic relationships commonly observed in 3D RNA structures and therefore reliably predict the quality of RNA structures. We started with the preparation of a high-quality training dataset based on experimentally determined structures retrieved from a non-redundant 3D RNA structures repository. Due to the scarcity of the available structures, we had to extend the dataset by 3D RNA models predicted using all state-of-the-art methods for 3D RNA structure prediction. We confirmed that graph NNs can harness a wide range of conformational space. We also solved the problem of variable-length 3D RNA structure representation crucial for the proper application of machine learning techniques by constructing local 3D motifs and applying an ensemble of local quality scores to assess the whole 3D RNA structure quality. A novel general quality assessment approach that does not need the reference 3D structure to reliably rank RNA 3D structures, we believe will be a breakthrough in the field of RNA structural bioinformatics.
Keywords
- Computational Biology, Bioinformatics and Medicine
- Artificial Intelligence
- Expert Systems and Neural Networks
Status: accepted
Back to the list of papers