New generative model predicts enzyme reaction mechanisms straight from sequence
KevinKaichuang · x · 2026-10-28
Ariane Mora's team presents a model that generates an enzyme's reaction mechanism from its sequence, catalytic residues, substrates and cofactors. Per the shared framework diagram, it casts function prediction as a generative problem, using sequence embeddings plus a bond-electron matrix fed into a Graph GPS (MPNN + Transformer) architecture to output product distributions, reactivity and activity. Training spans USPTO, M-CSA and EnzymeMap, with mechanism-similarity holdouts testing whether the model learns general biochemistry rules and generalizes to novel chemistry.
Related event: New Model Generates Enzyme Reaction Mechanisms Directly from Sequence(3 posts)→
More from Research
- Dev trains a 21B MoE (3.36B active) from scratch on 70B tokens and logs the emergence — Prestigious-Taste-63 · 2026-10-28
- LALA Bench: 236 Academic Questions Expose LLMs Answering from Memory, Not Reading — JasonBotterill · 2026-10-28
- RSS 2026 Area Chair hopes new review format will curb AI slop papers — DJiafei · 2026-10-28
- Amazon's EMMA treats multi-agent configs as episodic memory, beating baselines by up to 32.5% — omarsar0 · 2026-10-28
- NTU's MIDI-REPA uses MIDI alignment to improve text-to-music, trainable on one RTX 3090 in ~31h — affige_yang · 2026-10-28
- Google's EmbeddingGemma 2: one open embedding space for text, image, audio, video, setting a new Pareto frontier — kalyan_kpl · 2026-10-28