English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Deep learning has been a prevalence in computational chemistry and widelyimplemented in molecule property predictions. Recently, self-supervisedlearning (SSL), especially contrastive learning (CL), gathers growing attentionfor the potential to learn molecular representations that generalize to thegigantic chemical space. Unlike supervised learning, SSL can directly leveragelarge unlabeled data, which greatly reduces the effort to acquire molecularproperty labels through costly and time-consuming simulations or experiments.However, most molecular SSL methods borrow the insights from the machinelearning community but neglect the unique cheminformatics (e.g., molecularfingerprints) and multi-level graphical structures (e.g., functional groups) ofmolecules. In this work, we propose iMolCLR: improvement of MolecularContrastive Learning of Representations with graph neural networks (GNNs) intwo aspects, (1) mitigating faulty negative contrastive instances viaconsidering cheminformatics similarities between molecule pairs; (2)fragment-level contrasting between intra- and inter-molecule substructuresdecomposed from molecules. Experiments have shown that the proposed strategiessignificantly improve the performance of GNN models on various challengingmolecular property predictions. In comparison to the previous CL framework,iMolCLR demonstrates an averaged 1.3% improvement of ROC-AUC on 7classification benchmarks and an averaged 4.8% decrease of the error on 5regression benchmarks. On most benchmarks, the generic GNN pre-trained byiMolCLR rivals or even surpasses supervised learning models with sophisticatedarchitecture designs and engineered features. Further investigationsdemonstrate that representations learned through iMolCLR intrinsically embedscaffolds and functional groups that can reason molecule similarities.

Improving Molecular Contrastive Learning via Faulty Negative Mitigation and Decomposed Fragment Contrast