PlasticEnz: An integrated database and screening tool combining homology and machine learning to identify plastic-degrading enzymes in meta-omics datasets

TL;DR

PlasticEnz is a tool that helps find enzymes capable of breaking down plastic in environmental samples using a mix of sequence similarity and machine learning.

Contribution

PlasticEnz introduces a novel integration of homology-based search and machine learning for identifying plastic-degrading enzymes in metagenomic data.

Findings

PlasticEnz successfully identified known PETases and PHBases in both lab and real-world samples.

The tool distinguished between plastic-contaminated and pristine environments effectively.

PlasticEnz achieved an F1 score > 0.7 on an independent test set for PET and PHB classification.

Abstract

PlasticEnz is a new open-source tool for detecting plastic-degrading enzymes (plastizymes) in metagenomic data by combining sequence homology-based search with machine learning techniques. It integrates custom Hidden Markov Models, DIAMOND alignments, and polymer-specific classifiers trained on ProtBERT embeddings to identify candidate depolymerases from user-provided contigs, genomes, or protein sequences. PlasticEnz supports 11 plastic polymers with ML classifiers for PET and PHB, achieving F1 > 0.7 on an independent test set. Applied to plastic-exposed microcosms and field metagenomes, the tool recovered known PETases and PHBases, distinguished plastic-contaminated from pristine environments, and clustered predictions with validated reference enzymes. PlasticEnz is fast, scalable, and user-friendly, providing a robust framework for exploring microbial plastic degradation potential in complex communities.

Plastic pollution is a global problem, and one promising solution is to apply microbes to break them down. However, finding the enzymes responsible for this in complex environmental samples is not easy. We developed PlasticEnz, a free and easy-to-use tool that helps researchers identify plastic-degrading enzymes or “plastizymes” in metagenomic data. PlasticEnz combines traditional sequence similarity search methods with machine learning models trained on previously known plastizymes. It works with protein sequences, contigs, or genomes with ML components optimised for classification of two common plastizymes: PETases and PHBases. We tested PlasticEnz on both controlled lab experiments and real-world samples from plastic-polluted soils and clean environments. The tool successfully identified known plastic-degrading enzymes and even helped distinguish between polluted and pristine sites. By making plastizyme detection more accessible, PlasticEnz enables researchers to better explore the microbial potential for plastic degradation, which could support future bioremediation efforts.

Figures