GP-MoLFormer-Sim: Test Time Molecular Optimization through Contextual Similarity Guidance

TL;DR

This paper presents a training-free, test-time method for molecular optimization that guides a generative language model to produce molecules similar to a target, improving property optimization and drug design tasks.

Contribution

Introduces GP-MoLFormer-Sim, a novel test-time approach that uses contextual similarity to steer molecular generation without retraining the model.

Findings

Outperforms existing training-free methods on standard benchmarks.

Effective in property optimization, molecular rediscovery, and drug design.

Enhances generative control via similarity guidance in CLMs.

Abstract

The ability to design molecules while preserving similarity to a target molecule and/or property is crucial for various applications in drug discovery, chemical design, and biology. We introduce in this paper an efficient training-free method for navigating and sampling from the molecular space with a generative Chemical Language Model (CLM), while using the molecular similarity to the target as a guide. Our method leverages the contextual representations learned from the CLM itself to estimate the molecular similarity, which is then used to adjust the autoregressive sampling strategy of the CLM. At each step of the decoding process, the method tracks the distance of the current generations from the target and updates the logits to encourage the preservation of similarity in generations. We implement the method using a recently proposed $\sim$47M parameter SMILES-based CLM, GP-MoLFormer, and therefore refer to the method as GP-MoLFormer-Sim, which enables a test-time update of the deep generative policy to reflect the contextual similarity to a set of guide molecules. The method is further integrated into a genetic algorithm (GA) and tested on a set of standard molecular optimization benchmarks involving property optimization, molecular rediscovery, and structure-based drug design. Results show that, GP-MoLFormer-Sim, combined with GA (GP-MoLFormer-Sim+GA) outperforms existing training-free baseline methods, when the oracle remains black-box. The findings in this work are a step forward in understanding and guiding the generative mechanisms of CLMs.