STAMP as a topology-aware molecular language for fragment-level learning and editing
A fragment-tree molecular language that makes motifs, attachment sites, and topology explicit for representation learning and executable editing.
Citation
Yuxuan Ren, Jiatong Li, Ziqiao Meng, and Yatao Bian. (2026). STAMP as a topology-aware molecular language for fragment-level learning and editing. arXiv:2610.18438.
About this work
STAMP stands for Sequential Tree Attachment-based Motif-Preserving representation. This molecular language encodes motif identity, attachment sites, and fragment-tree topology. It provides reusable chemical fragments and addressable subtrees for structural editing.
Across molecular generation, text-to-molecule translation, property prediction, activity-cliff prediction, optimization, and instruction-guided editing, STAMP connects learned fragment representations with executable operations. Its deterministic compiler also allows frozen language models to act on indexed molecular fragments instead of rewriting complete molecular strings.