Evaluating the Progression of Large Language Model Capabilities for Small-Molecule Drug Design

TL;DR

This paper introduces chemically-grounded tasks and RL-based post-training to evaluate and improve large language models for small-molecule drug design, highlighting significant potential and current limitations.

Contribution

It presents a new suite of chemically-grounded tasks as RL environments and demonstrates how targeted post-training enhances LLM performance in drug discovery.

Findings

Frontier models excel at chemical tasks but have room for improvement.

RL-based post-training significantly boosts model performance.

A smaller model with targeted training rivals state-of-the-art models.

Abstract

Large Language Models (LLMs) have the potential to accelerate small molecule drug design due to their ability to reason about information from diverse sources and formats. However, their practical utility remains unclear due to the lack of benchmarks that reflect real-world scenarios. In this work, we introduce a suite of chemically-grounded tasks spanning molecular property prediction, molecular representation transformations, and molecular design. Importantly, we formulate these tasks as reinforcement learning (RL) environments, enabling a unified approach for evaluation and post-training. Across three model families, we find that frontier models are increasingly proficient at chemical tasks, but that there is significant room for improvement, especially in experimental settings with low data. Critically, we show that RL-based post-training can substantially improve performance. A smaller model post-trained on our environments becomes competitive with state-of-the-art frontier models, despite a significantly weaker base model. This suggests a practical route toward employing LLMs in drug discovery; by combining carefully-designed evaluation tasks with targeted post-training, we can both elucidate and close critical capability gaps.