<ul><li>A new method called GP-MoLFormer-Sim has been introduced for molecular optimization by leveraging contextual similarity guidance.</li><li>The method uses a generative Chemical Language Model (CLM) to navigate and sample from the molecular space while preserving similarity to a target molecule.</li><li>GP-MoLFormer-Sim adjusts the autoregressive sampling strategy based on molecular similarity estimates to maintain similarity in generated molecules.</li><li>The method, combined with a genetic algorithm (GA), outperforms existing training-free baseline methods in various molecular optimization benchmarks.</li></ul>

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

Discover more