2021/06/03 by Rémy Brossard, Brossard, Rémy, Oriel Frigo +3
Biochemistry, Genetics and Molecular Biology · Computer Science · Materials Science · Mathematics · Physics and Astronomy · #Algorithm #Artificial intelligence #Chemical Physics (physics.chem-ph) #Computational Drug Discovery Methods #Computer science #FOS: Computer and information sciences #FOS: Physical sciences #Graph #Machine Learning (cs.LG) #Machine Learning in Materials Science #Machine learning #Mathematical optimization #Mathematics #Optimization problem #Protein Structure and Dynamics #Sampling (signal processing) #Similarity (geometry) #Theoretical computer science #cs.LG #physics.chem-ph
paper · pdf · doi:10.48550/arxiv.2106.13318
published in arXiv (Cornell University) (Cornell University) · 15 pages (9 page main article without refs or appendix) and 2 figures. In review at NEURIPS 2021
arxiv created 2021/06/03 · openalex publication_date 2021/06/03 · arxiv updated 2021/06/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Deep learning based molecular graph generation and optimization has recently been attracting attention due to its great potential for de novo drug design. On the one hand, recent models are able to efficiently learn a given graph distribution, and many approaches have proven very effective to produce a molecule that maximizes a given score. On the other hand, it was shown by previous studies that generated optimized molecules are often unrealistic, even with the inclusion of mechanics to enforce similarity to a dataset of real drug molecules. In this work we use a hybrid approach, where the dataset distribution is learned using an autoregressive model while the score optimization is done using the Metropolis algorithm, biased toward the learned distribution. We show that the resulting method, that we call learned realism sampling (LRS), produces empirically more realistic molecules and outperforms all recent baselines in the task of molecule optimization with similarity constraints.