chaohua06/EnzSub-dataset

EnzSub Training and Tm Data

Summary

This repository contains the author-assembled or author-processed data released with EnzSub. It includes the enzyme sequences used for continued pretraining (CPT), the processed enzyme-substrate tables used for substrate-aware training (SUB), and the fixed train/validation/test data used for the melting-temperature (Tm) downstream task.

Other downstream and transfer benchmarks used by EnzSub originate from external projects and are intentionally not redistributed here. Obtain those datasets from their original repositories and follow the citations in the EnzSub paper and code repository.

Code and reproducibility instructions are available at AGI-FBHC/EnzSub.

Directory structure

cpt/
  ENZ30_train.fasta
  ENZ30_val.fasta
sub/
  oed.csv
  oed_enzyme_stats.csv
tm/
  train.csv
  val.csv
  test.csv
  train.fasta
  val.fasta
  test.fasta
checksums.sha256

File inventory

File	Records	Description
cpt/ENZ30_train.fasta	1,022,914	ENZ30 CPT training sequences
cpt/ENZ30_val.fasta	10,000	ENZ30 CPT validation sequences
sub/oed.csv	17,058	Processed enzyme-substrate pair table
sub/oed_enzyme_stats.csv	6,438	Per-enzyme substrate descriptor statistics
tm/train.csv	5,972	Fixed Tm training split
tm/val.csv	814	Fixed Tm validation split
tm/test.csv	932	Fixed Tm test split
tm/train.fasta	5,972	FASTA representation of the Tm training split
tm/val.fasta	814	FASTA representation of the Tm validation split
tm/test.fasta	932	FASTA representation of the Tm test split

checksums.sha256 records the SHA-256 digest of every released data file.

Formats

The ENZ30 files use standard FASTA records with source sequence identifiers and descriptions in the headers. The Tm FASTA headers contain the internal sequence identifier, UniProt identifier, gene name when available, melting temperature, and cluster identifier.

The main fields in sub/oed.csv are:

• UNIPROT, Sequence: enzyme identifier and amino-acid sequence;
• substrate_key, SUBSTRATE, SMILES: substrate identifiers and structure;
• EC, ec_major_classes: enzyme commission annotation;
• scaffold_smiles: molecular scaffold;
• sample_weight, confidence_tier, remove_category: processing and quality fields;
• molecular descriptor columns including molecular weight, LogP, TPSA, hydrogen-bond counts, ring counts, and fraction Csp3.

sub/oed_enzyme_stats.csv contains per-enzyme counts and the mean and standard deviation of the corresponding substrate descriptors.

Each Tm CSV contains 11 columns: sequence, UniProt and gene identifiers, meltPoint, sequence length, EC number, internal sequence identifier, cluster identifier, cluster size, and the fixed split label. Melting temperature is reported in degrees Celsius.

Use with EnzSub

Download the dataset into the code repository and retain these directory names:

hf download chaohua06/EnzSub-dataset \
  --repo-type dataset \
  --local-dir data

The resulting paths match the public EnzSub configurations:

data/cpt/ENZ30_train.fasta
data/cpt/ENZ30_val.fasta
data/sub/oed.csv
data/sub/oed_enzyme_stats.csv
data/tm/train.csv
data/tm/val.csv
data/tm/test.csv

The Tm CSV files are sufficient for the released-embedding reproduction. The FASTA copies are provided for sequence inspection and workflows that regenerate representations.

Provenance and scope

These files are the fixed processed inputs used by the released EnzSub configurations. Raw construction notebooks, exploratory intermediates, caches, model embeddings, and generated results are outside this dataset repository. Precomputed representations are available separately at chaohua06/EnzSub-Embeddings.

Access and license

No license has yet been assigned to this dataset. Public access does not by itself grant permission to reuse, modify, or redistribute the files. The rights statement will be updated after the authors complete their licensing review.

Citation

The preferred citation and article DOI will be added after publication. Until then, cite the EnzSub GitHub repository and the exact Hugging Face dataset revision used in an analysis.