petadex/petai-halo-screen

PETai Halo Screen

An in-house BHET plate-clearing screen of 253 candidate genes, run by iGEM Toronto 2026. 197 of them resolve to a sequence and carry a binary label: 29 active, 168 inactive. This is the halo source of petadex/petai-activity-labels and it supplies 168 of that release's 200 negatives.

Unlike the other three sources, nothing here was read out of a paper. These are our own measurements, published with the plate-level data behind them.

Why a screen of mostly negatives is the useful part

Published activity data is almost entirely positive: a paper reports the enzyme that worked. Across the two literature sources in this release, 8,000 or so numeric labels come with 32 negatives. A screen reports what did not clear the plate as readily as what did, so 168 of these 197 labels are negatives, and they are the reason a binary classifier trained on the unified release has two classes at all.

Files

File	Rows	What it is
labels.csv	197	One row per gene with a sequence, at the label condition
conditions.csv	3,036	Every gene at every condition: 253 genes, 2 media, 6 timepoints
plates.tsv	6,792	Plate-level colony measurements, the raw readout
sequences.fasta	199	The screened sequences, named by MD5. 197 carry a label
manifest.json		Counts and input checksums

source_row_id is the gene identifier, which is its database accession, for example ADV92526.

The label

One condition is promoted to the label: BHET 25 at 120 hours, active where z > 3.0, z being the clearing score against an empty-vector control on the same plate set. conditions.csv carries all 12 condition combinations with the promoted one flagged is_label_condition, so a different threshold, medium or timepoint can be used instead, and plates.tsv holds the colony measurements the z scores were computed from.

labels.csv schema

source_row_id, seq_md5, task (binary), variable (plate clearing), unit (z vs empty vector), substrate (BHET), value (the z score), is_negative, ph and temperature_c (empty), time_h (120), ref (halo:BHET25@120h), curator (wet-lab), n_replicates, z.

Read before using these labels

• BHET is not PET. This is a soluble dimer assay in a plate. A negative means no detectable BHET hydrolysis under these conditions, not that the enzyme is not a PETase. A positive means the opposite of nothing about polymer activity.
• Heterologous expression confounds every negative. A gene that failed to express, folded badly or was toxic in the host scores as inactive. Some fraction of the 168 negatives is expression failure rather than catalytic failure, and this screen cannot tell you which.
• Replication is thin. 179 of 197 genes have 2 replicates and 18 have 4.
• z is not a graded activity measure. It runs from -1651.69 to 240.29 with a median of -13.14. The large negative values reflect colony growth differences rather than degrees of inactivity. Use the binary call, or re-derive your own from plates.tsv.
• 56 of the 253 screened genes have no sequence and so carry no label. They are present in conditions.csv for completeness.
• The panel is not a random sample. Genes were picked as plausible candidates, so the active rate here says nothing about the base rate in a mined corpus.

Licensing

CC BY 4.0. This screen is entirely our own work, so the license applies cleanly to the measurements, the z scores, the plate data and the compilation. CC BY 4.0 also licenses the database right where that right exists.

The one element that is not ours is the amino acid sequences of the screened genes, which came from public sequence databases and carry those databases' terms. Sequence data itself is not a copyrightable work.

Attribution request, beyond what the license obliges: these are 197 wet lab results produced by an undergraduate team over one season. If they are the negatives that make your classifier trainable, say so.

Citation

@misc{petai_halo_screen_2026,
  title  = {PETai Halo Screen},
  author = {{iGEM Toronto 2026}},
  year   = {2026},
  url    = {https://huggingface.co/datasets/petadex/petai-halo-screen}
}

Staged by build_source_repos.py from the normalized halo assay tables.