AlphaMol: rigid neighborhood representation for small molecule structure prediction
Keywords: molecule, deep learning, structure prediction, chemistry, biology, biomolecules, bioinformatics
TL;DR: A new representation of molecules as a collection of neighborhoods that behave as rigid bodies and a way to encode the bonds between rigid bodies into a prediction algorithm.
Abstract: Recent success of the deep learning-based approach AlphaFold2 revolutionized the field of protein structure prediction. Since AlphaFold2 development, a lot of efforts were made to shape its limitations as well as to improve it further with respect to difficult protein classes. However, structure prediction of non-protein type of molecules, such as small organic molecules, non-standard amino acids, nucleic acids, and others, is still an open problem. Inspired by the powerful AlphaFold2 neural network architecture, we developed a general framework for prediction of molecular structures of arbitrary type. Specifically, we developed novel representation of molecular structures as a collection of rigid-body neighborhoods with encoded bonds between the neighborhoods fed into neural network comprising developed Evoformer-like blocks. We tested our approach on small organic molecules, that possess much higher variability in terms of atomic composition and structural patterns compared to proteins. Namely, we applied the developed method, named AlphaMol, to the ground-state structure prediction problem of small molecules and observed superior performance metrics on the PubChemQC benchmark, compared to the existing approaches. Our results demonstrate possibility to create multi-modal molecular structure prediction methods, that operate across different molecular types. The AlphaMol source code is available in the repository: https://anonymous.4open.science/r/AlphaMol-2EA7.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4380
Loading