Score-based Diffusion Model for Conformer Generation.

Molecule conformer generation is an important task in several scientific fields, such as bioinformatics, pharmacology, and material discovery, to name a few, that aims to construct the 3D structure of a molecule. Many properties of a molecule are determined by its 3D structure. Normally, we can do experiments to determine a range of structures of a molecule. However, conformers are not available in certain circumstances, partly because of limited resources. As a result, alternate techniques for building 3D buildings are critical. In drug discovery, conformers of the molecule can be generated by computational means. In the past decades, a large number of conformer generation approaches have been developed for molecules. These procedures, however, are time-consuming and produce a large number of conformers. Recently, machine learning has come into play as a computational tool for accelerating the process of conformer generation with high-quality samples. This paper shows our research on a diffusion model for generating conformers.