Deep learning has become ubiquitous in drug discovery, providing significant acceleration and cost reduction. As the most fundamental building block, molecular representation is essential for predicting molecular properties and activities to enable various downstream applications in drug discovery. While striving to incorporate diverse sources of information for comprehensive representation, current approaches can inadvertently introduce noise that impedes predictions, especially when labeled data are scarce. To address this, we propose a novel approach, MolTailor, which shifts the focus from incorporating more information to selectively optimizing representation by accentuating task-relevant features. Specifically, MolTailor builds upon understanding natural language descriptions of tasks, and tailors molecular representation by prioritizing the most pertinent features. Evaluations demonstrate MolTailor's superior performance over baselines, validating the efficacy of enhancing relevance for molecular representation learning. This illustrates the potential of language model guided optimization to better exploit and unleash the capabilities of existing informative molecular representation.

Haoqiang Guo

1

SHORT BIO

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES