ZymPro

ZymPro dataset, ranking split update, September 25, 2026. Transition-state generation: 28,974 / 3,141 / 3,142 train/validation/test records. Enzyme compatibility ranking: 87,849 / 6,464 / 5,255 groups, spanning 178 / 84 / 75 distinct reaction rules. Product generation: 110,646 / 10,313 / 9,519 records. The ranking validation and test subsets apply the rule exclusions documented in RANKING_SPLIT_UPDATE.json, including records with excluded rules in multi-rule combinations. Training membership and the other task splits are retained. The accompanying CSV contains metrics recomputed from saved predictions on the retained ranking test groups. The release preserves the original repository paths across 12 category ZIP archives. Four large category ZIPs are transported as consecutive 400 MB byte parts. Concatenate each numbered part set to reconstruct its ZIP, then extract all 12 ZIPs into the same repository root. README-category-archives.md and manifest.json list the assembly instructions, files, sizes, and checksums.