An evaluation of unconditional 3D molecular generation methods

TL;DR

This paper evaluates recent unconditional 3D molecular generation methods, revealing that SemlaFlow outperforms others with high validity, uniqueness, and novelty rates, prompting a re-evaluation of current benchmarks.

Contribution

It provides a comprehensive comparison of five recent 3D molecular generation methods, highlighting the superior performance of SemlaFlow and emphasizing the need to update evaluation benchmarks.

Findings

SemlaFlow achieves 87% success rate without post-processing.

Post-processing increases success rate to 92.4%.

Current benchmarks may be saturated, requiring re-evaluation.

Abstract

Unconditional molecular generation is a stepping stone for conditional molecular generation, which is important in \emph{de novo} drug design. Recent unconditional 3D molecular generation methods report saturated benchmarks, suggesting it is time to re-evaluate our benchmarks and compare the latest models. We assess five recent high-performing 3D molecular generation methods (EQGAT-diff, FlowMol, GCDM, GeoLDM, and SemlaFlow), in terms of both standard benchmarks and chemical and physical validity. Overall, the best method, SemlaFlow, has a success rate of 87% in generating valid, unique, and novel molecules without post-processing and 92.4% with post-processing.