Utilizing Perturbation of Atoms' Positions for Equivariant Pre-Training in 3D Molecular Analysis

Tal Kiani, Avi Caciularu, Shani Zev, Dan Thomas Major, Jacob Goldberger

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Over the past few years, a number of Graph Neural Network (GNN) architectures have been effectively employed for molecular analysis. However, generating annotated molecular data usually requires molecular dynamics or quantum chemistry calculations, which can be extremely time-consuming. To address this challenge, we introduce a predictive equivariant self-supervision technique that is founded on perturbing the 3D positions of the atoms. This method is ideal for 3D molecular data and allows the network to initially learn general structural information before fine-tuning it for specific tasks. We demonstrate that these pre-training procedures can also be utilized to fine-tune the network for learning molecular properties on a different dataset. Our pre-training method is demonstrated to surpass previously proposed solutions via extensive experiments on different standard molecular datasets.

Original languageEnglish
Title of host publicationProceedings of the 2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing, MLSP 2023
EditorsDanilo Comminiello, Michele Scarpiniti
PublisherIEEE Computer Society
ISBN (Electronic)9798350324112
DOIs
StatePublished - 2023
Event33rd IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2023 - Rome, Italy
Duration: 17 Sep 202320 Sep 2023

Publication series

NameIEEE International Workshop on Machine Learning for Signal Processing, MLSP
Volume2023-September
ISSN (Print)2161-0363
ISSN (Electronic)2161-0371

Conference

Conference33rd IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2023
Country/TerritoryItaly
CityRome
Period17/09/2320/09/23

Bibliographical note

Publisher Copyright:
© 2023 IEEE.

Keywords

  • 3D Molecular Analysis
  • Equivariance
  • Graph Neural Networks
  • pre-training

Fingerprint

Dive into the research topics of 'Utilizing Perturbation of Atoms' Positions for Equivariant Pre-Training in 3D Molecular Analysis'. Together they form a unique fingerprint.

Cite this