Saved in:
Bibliographic Details
Main Authors: Antoniuk, Evan R., Li, Peggy, Keilbart, Nathan, Weitzner, Stephen, Kailkhura, Bhavya, Hiszpanski, Anna M.
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.02059
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929659568979968
author Antoniuk, Evan R.
Li, Peggy
Keilbart, Nathan
Weitzner, Stephen
Kailkhura, Bhavya
Hiszpanski, Anna M.
author_facet Antoniuk, Evan R.
Li, Peggy
Keilbart, Nathan
Weitzner, Stephen
Kailkhura, Bhavya
Hiszpanski, Anna M.
contents Although generative models hold promise for discovering molecules with optimized desired properties, they often fail to suggest synthesizable molecules that improve upon the known molecules seen in training. We find that a key limitation is not in the molecule generation process itself, but in the poor generalization capabilities of molecular property predictors. We tackle this challenge by creating an active-learning, closed-loop molecule generation pipeline, whereby molecular generative models are iteratively refined on feedback from quantum chemical simulations to improve generalization to new chemical space. Compared against other generative model approaches, only our active learning approach generates molecules with properties that extrapolate beyond the training data (reaching up to 0.44 standard deviations beyond the training data range) and out-of-distribution molecule classification accuracy is improved by 79%. By conditioning molecular generation on thermodynamic stability data from the active-learning loop, the proportion of stable molecules generated is 3.5x higher than the next-best model.
format Preprint
id arxiv_https___arxiv_org_abs_2501_02059
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Active Learning Enables Extrapolation in Molecular Generative Models
Antoniuk, Evan R.
Li, Peggy
Keilbart, Nathan
Weitzner, Stephen
Kailkhura, Bhavya
Hiszpanski, Anna M.
Machine Learning
Materials Science
Chemical Physics
Although generative models hold promise for discovering molecules with optimized desired properties, they often fail to suggest synthesizable molecules that improve upon the known molecules seen in training. We find that a key limitation is not in the molecule generation process itself, but in the poor generalization capabilities of molecular property predictors. We tackle this challenge by creating an active-learning, closed-loop molecule generation pipeline, whereby molecular generative models are iteratively refined on feedback from quantum chemical simulations to improve generalization to new chemical space. Compared against other generative model approaches, only our active learning approach generates molecules with properties that extrapolate beyond the training data (reaching up to 0.44 standard deviations beyond the training data range) and out-of-distribution molecule classification accuracy is improved by 79%. By conditioning molecular generation on thermodynamic stability data from the active-learning loop, the proportion of stable molecules generated is 3.5x higher than the next-best model.
title Active Learning Enables Extrapolation in Molecular Generative Models
topic Machine Learning
Materials Science
Chemical Physics
url https://arxiv.org/abs/2501.02059