This is the authors' abstract. We don't add key points for this paper.
Molecular graphs and molecular strings are two well-established molecular representations for deep learning, with complementary strengths: while graphs make atom and bond properties explicit, strings are sequential and straightforward to generate. Here, we introduce a novel molecular representation than bridges these two representations: Multichannel Atom and Topology for Chemical Representation Series (MATCHERS). In a single representation, MATCHERS encode user-defined atomic properties, explicit molecular topology, as well as bond properties, while maintaining the sequential character typical of molecular strings. In this proof-of-concept, we show that MATCHERS outperform sequential string representations (SMILES and SELFIES) for bioactivity prediction, and that they can be used for de novo molecular design. MATCHERS illustrate the untapped opportunity of explicitly encoding chemical knowledge into new molecular representations that remain compatible with deep learning.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Brinkmann et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: