Neural Petri flows for chemical reactions
Neural Petri Flow predicts chemical reactions by learning rates while hard-wiring Petri-net conservation rules.
Neural Petri Flow is a neural architecture that remains a Petri net for every weight setting by learning only the transition rate law and fixing conservation and enabling rules as parameter-free layers. On valence nets, atom mapping, reaction classification, and forward prediction share one firing vector. Without training, the minimum firing vector maps 88.8% of the curated Golden set versus 85.6% for RXNMapper, and 88.7% versus 77.9% of EnzymeMap enzymatic reactions. Trained on USPTO-480K firing vectors it predicts 87.7% of products, and 90.2% of ECREACT EC numbers at the third level.
- Neural Petri Flow learns only the reaction rate law and hard-wires Petri-net semantics.
- Untrained minimum firing maps 88.8% of the Golden set versus 85.6% for RXNMapper.
- On USPTO-480K it predicts 87.7% of products, and 67.4% from a 1% subset.
- EC numbers on ECREACT reach 90.2% at the third level, 5.6 points above the prior best.
- AI models
- Neural Petri Flow
From the original article
1 min read at arxiv.orgPetri nets have been used to describe chemical processes such as reactions. They map well to chemistry: Places are the bonds between atoms and the free valence of each atom, a token is a unit of bond order, a transition forms or breaks a bond, the conserved quantities are the valence budgets of the atoms, and the enabling rule is the valence rule.
By Jose Eduardo Escrig Molina · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.