petadex/petai-labels-kinetics

PETai Labels, Published Kinetics

1,342 kinetic labels (kcat 454, Km 454, kcat/Km 434) on 206 enzyme sequences, compiled by hand from 74 papers. This is the kinetics source of petadex/petai-activity-labels and the table behind the /kinetics page on petadex.net.

source_row_id is the Experiment_# of the source row, so every unified label resolves to one row of plastic_kinetics_published.csv.

Start here rather than with the other sources

Every row of the source table carries its own amino acid sequence, so there is no link step and nothing to lose to a missing sequence. Conditions are well covered too: temperature on 78.1% of labels and pH on 71.2%, against 42.8% and 90.4% in the human-curated set and nothing at all in the LLM set. The target is a real kinetic constant rather than an assay proxy.

Only 14 of its sequences also appear in the human-curated set, so the two are complementary.

Files

File	Rows	What it is
labels.csv	1,342	Long format, one row per parsed constant
plastic_kinetics_published.csv	464	The source table verbatim, 18 columns
sequences.fasta	208	The distinct sequences, KIN0001 onward, headed with the enzyme names that share each sequence
manifest.json		Counts and input checksums

206 of the 208 sequences carry a label. The two that do not belong to the dead variants described below.

Why 464 rows become 1,342 labels

Each source row holds up to three constants, and one label is emitted per constant that parses. kcat and Km parse on 454 rows each and kcat/Km on 434; the rest are ranges, inequalities or blanks. The 10 rows that yield nothing are the dead variants.

labels.csv schema

source_row_id, seq_md5, task (quantitative throughout), variable (kcat, Km, kcat/Km), unit (1/s, M, 1/s/M), substrate, value, is_negative (false throughout), ph, temperature_c, time_h (empty throughout), ref (DOI), curator (empty, this table predates per-curator attribution).

Read before training

• Model log10 of the value. kcat/Km spans orders of magnitude.
• The three constants are three targets, not one. Do not pool them.
• Most rows measure a surrogate, not a plastic. 328 of 464 rows use a soluble ester: p-NP-C2 through p-NP-C18 on 245 rows, acetylcholine on 50, plus naphthyl esters, vinyl esters and triglycerides. Only 64 rows measure a plastic or a plastic oligomer: BHET 69 labels, PBSA 60, MHET 33, PET-02mer. If the task is PET hydrolysis, treat the surrogate rows as a related task to transfer from rather than as PET labels.
• SORT is undocumented. The Class column covers substrate class plus variant status in 13 values, and 8.0-SORT_WT plus 8.1-SORT_MUT account for 222 rows, or 48% of the table. What SORT denotes is not recorded anywhere we can find. Ask before folding those rows into a PET-specific target.
• Hold out sequence clusters, not rows. One enzyme contributes many rows across substrates and conditions.
• All 18 source columns are stored as text, numbers included. Cast before use. labels.csv is already typed.

The dead variants

10 source rows are catalytically dead point mutants, for example IsPETase-S160A-(DEAD), classed 0.2-PET_Dead. They are the only candidate negatives in this table and the cleanest in principle, since a dead active site is an unambiguous negative rather than a detection limit. They are unusable as published: the kcat, Km and kcat/Km cells are all blank and the rows carry no DOI, so they produce no label and cannot be traced to a paper. Filling in the value and the reference for those 10 rows would be a cheap and real improvement to the negative supply across the whole release.

Two defects in the SMILES, confirmed in this export

• Every pNP SMILES is malformed. All 13 encode the nitro group as [N+](O)=O, for example CCCC(=O)OC1=CC=C(C=C1)[N+](O)=O for p-NP-C4. The correct form is [N+](=O)[O-]. As written each carries a spurious +1 charge and an extra hydrogen, so a cheminformatics toolkit will either reject them or compute wrong descriptors without complaining. Fix the nitro encoding before using SMILES as features.
• 2-(1-Naphthyl)butanoic acid carries the SMILES for 1-naphthyl butanoate, a different compound and the same one as alpha-Naphthyl butyrate. Six rows are affected.

The BHET, MHET, PET-02mer and PBSA SMILES were spot checked and look correct.

Licensing

CC BY 4.0, which covers what is ours to license: the compilation, the parsing and the derived columns. CC BY 4.0 also licenses the database right in the compilation where that right exists.

What it does not cover:

• The constants are facts reported in the cited papers. Cite the DOI in ref when you use a value, not only this dataset.
• The papers' own text, tables and figures stay with their publishers. No source PDF is redistributed here.
• Sequences and accessions resolved against UniProt, GenBank and RCSB carry those services' own terms. UniProt is CC BY 4.0.

Citation

@misc{petai_labels_kinetics_2026,
  title  = {PETai Labels, Published Kinetics},
  author = {{iGEM Toronto 2026}},
  year   = {2026},
  url    = {https://huggingface.co/datasets/petadex/petai-labels-kinetics}
}

Exported from public.plastic_kinetics_published by export_kinetics.mjs and staged by build_source_repos.py.