# AlphaEnzyme raw data

This archive contains the raw tables and the small released input assets used
by the three tasks: specificity reference banks, reaction fingerprints,
kinetics descriptor banks and the eight mutation fold-normalization files.

Unpack into `data/` so that the paths are:

```text
data/inputs/specificity/tables/{train,valid}.csv
data/inputs/specificity/eval/eval.csv
data/inputs/specificity/features/reference/positive_context.npz
data/inputs/specificity/features/reference/positive_protein_pocket.npz
data/inputs/specificity/features/drfp/reaction_fingerprints.pkl
data/inputs/kinetics/{train,valid,test}.csv
data/inputs/mutation/{variants,references,proteingym_rows}.csv
data/inputs/mutation/{folds,protocol}.json
data/inputs/mutation/component_inputs_unlabelled.csv
data/inputs/mutation/inputs/rows.csv
data/inputs/mutation/inputs/fold_*_{train,valid,test}_{rows,indices}.*
data/inputs/mutation/features/fold_*_normalization.npz
```

The large per-pair Pairformer/ESMC feature caches are not included. The mutation
`inputs/` directory contains the released fold row tables and index arrays. See
`docs/NEW_DATA.md` for the feature-generation commands.

`CONTRACT.json` lists the machine-readable layout.