Revisiting molecular descriptors with TDiMS for interpretable intramolecular interactions based on substructure pairs.

TDiMS descriptor outperforms existing methods by accurately capturing nonlocal intramolecular interactions with 20% higher predictive accuracy in property prediction tasks.

• Why it matters: Understanding long-distance structural features is vital for predicting molecular properties, yet current descriptors lack effective ways to incorporate these nonlocal interactions, limiting their interpretability and accuracy.
• What they did: The authors developed TDiMS, a descriptor that summarizes pairwise topological distances between molecular substructures, evaluated on large molecules and property prediction benchmarks.
• The result: TDiMS not only surpasses other descriptors in predictive performance but also provides highly interpretable features, enabling better insights for materials discovery and molecular design.