PANTS: a triaged catalogue of candidate PET-degrading enzymes for therapeutic conditions

PANTS: PETase ANnotation and Triage System. Candidate polyesterases mined from metagenomic sequence space and triaged for therapeutic use: degradation of PET at 37 °C and neutral pH rather than in an industrial reactor above PET's glass transition. 14,804,920 predicted proteins were scanned across four environments and 439 candidates retained, each carrying its retrieval scores, catalytic triad positions found by geometry rather than by motif, and active-site measurements from a predicted structure. human gut: 311 candidates from 12,584,458 proteins compost: 69 candidates from 1,020,575 proteins marine plastisphere: 44 candidates from 737,027 proteins landfill: 15 candidates from 436,229 proteins The distinction that matters most in this dataset: 333 of 846 reference positives carry a measurement; the rest carry annotation. Entries from PAZy are present because activity was measured on a plastic and published. Entries labelled EC 3.1.1.101 via ECO:0000256 were assigned by sequence similarity, and a model trained on those reaches AUC 1.000 by reproducing the annotation rule, which is circular. Trained on the measured set alone the activity head reaches AUC 0.976 across 45 independent sequence clusters, against 0.829 for a classifier using amino-acid composition and length only. Known limitations, stated up front. This is a snapshot taken before structure prediction finished: 268 candidates have a structure and 257 have geometry. 14 candidates over 450 aa were deliberately not folded, being probable fusions or misassemblies. There is no demonstrated discrimination of PET activity within the polyesterase family, and that question currently has no dataset, because databases record what works and the negative class is largely unwritten. Predicted structures are predictions. Nothing here addresses delivery, immunogenicity or the fate of liberated TPA and EG in vivo. See DATASHEET.md. Derived from UniProt and UniRef (CC BY 4.0), ESTHER, PAZy (Buchholz et al. 2022), MGnify assemblies, the PDB (CC0), and the human gut PET hydrolases of Zhang et al. 2024 (PMID 39551294). Code: github.com/bellcheddar/PANTS.