Back

EVODEX: A Mechanistic Framework for Extracting, Structuring, and Predicting Enzymatic Reactivity

Conceicao, L. L.; Lin, H.; Tai, D.; Xue, G.; Zhang, H.; Zhang, C.; Anderson, J. C.

2025-06-25 bioinformatics
10.1101/2025.06.19.660615 bioRxiv
Show abstract

Accurately modeling enzymatic reactivity is essential for synthesis planning, mass spectrometry interpretation, and biochemical knowledge extraction, but existing reaction datasets are often incomplete or inconsistently annotated. EVODEX introduces two abstractions to address this: partial reactions, which isolate atom-mapped substrate-product transformations from complex biochemical reactions, and Electronic Reaction Operators (EROs), which encode the bonding and orbital environment of atoms undergoing change. From 349,458 curated reactions in EnzymeMap, EVODEX constructs over 186,000 partial reactions and mines a core set of 1,404 EROs, each supported by at least 10 examples. Together, these explain 91 percent of test reactions at the formula level and 62 percent at the mechanistic level. A further filtered synthesis subset of 436 one-to-one operators that exclude ubiquitous metabolites provides a compact and generalizable vocabulary of single-step enzymatic transformations. EVODEX operators are auditable, reusable, and available as part of an open-source Python package with datasets, Jupyter notebooks, and a browsable website.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.