Preprint reframes enzyme function prediction as conditional generation with Fluxion
LeoTZ03 · x · 2026-10-27
A bioRxiv preprint by William Rieger et al. recasts enzyme function prediction as conditional generation: given an enzyme sequence, substrates, cofactors and minimal catalytic residues, the model (PLM encoder + Graph GPS/MPNN with cross-attention) generates predicted products with frequencies via a bond-electron matrix representation.
Training is staged across USPTO chemistry (2.17M reactions), M-CSA (2,747) and EnzymeMap (28,837), with held-out tests on mechanism-similar enzymes and unseen non-native chemistry to probe whether the model learns general rules of biochemistry. Output heads predict reactivity, activity and bond-electron changes, with SMILES filtering for invalid products. The author presents the work at the ML4Proteins seminar series.
More from Research
- Bengio et al. new paper: quantifying compositional structure via algorithmic information theory — patrickmineault · 2026-10-27
- 'Lessons learned' papers may be worth more to AI models than 2500-citation atlas papers — owl_posting · 2026-10-27
- AI reportedly disproves the Exponential Time Hypothesis, shifting P vs NP odds — esrtweet · 2026-10-27
- Adapter-LoRA on Evo2 + Qwen3 beats Carbon-X DNA model with 24h of cheap compute — danielmckinn0n · 2026-10-27
- "I Just Ran 2 Million Regressions" Pt. 2: Massively Parallel Empirical Analysis — paulnovosad · 2026-10-27
- Princeton's PaperScorer fine-tunes Qwen to predict paper acceptance, hitting 77.5% accuracy — IanArawjo · 2026-10-27