Oracle-Budgeted Molecular Optimization with Short-Term Graph Memory

2026-07-30

AI Summary

This work addresses the challenge of efficiently selecting high-potential candidate molecules under a limited oracle query budget in molecular optimization. The authors propose a plug-in short-term graph memory mechanism that, without altering the generator architecture, incrementally constructs a graph neural network surrogate model from previously evaluated molecules to prescreen candidates and prioritize those with higher predicted utility for oracle evaluation. This approach incurs no additional oracle calls yet significantly enhances optimization efficiency. Experimental results demonstrate that, under a stringent budget of only 1,000 oracle queries, the method consistently outperforms the original strategy across four fragment-based generators, substantially improving the average top-10 score without any observed performance degradation, while also revealing systematic interactions between the generator’s exploration–exploitation behavior and surrogate-guided selection.

Abstract

Molecular optimization is commonly performed under a limited oracle budget, which makes deciding what to evaluate as important as deciding what to generate. We introduce short-term graph memory, a plug-in module that preserves the generator architecture and native update rule while learning from previously evaluated molecules to prioritize subsequent oracle queries. The module maintains an online graph neural surrogate that pre-screens each round's candidate pool, so the fixed oracle budget is spent on molecules with higher predicted utility. Applied to a fragment-based generator on a standard molecular optimization benchmark, it improves the mean top-10 score at no extra oracle cost and never falls behind the base on any oracle; the gain extends to all four generators we tested at a tight budget of one thousand calls. We then analyze how surrogate-guided selection interacts with the exploration and exploitation behavior of different generators. Its benefit at larger budgets is consistent with two properties of the backbone: how broadly it searches, and how effectively its native search already exploits oracle feedback. We provide a simple way to spend a fixed oracle budget more selectively, and evidence on which generators benefit from it.

Problem

• molecular optimization
• oracle budget
• candidate selection
• graph neural networks
• surrogate model

Innovation

• short-term graph memory
• oracle-bounded optimization
• graph neural surrogate
• molecular optimization
• fragment-based generation