Discovering hidden candidate plastic-degrading enzymes: Combined multi-omics and machine learning strategy

Area of Science:

• Biotechnology and Environmental Microbiology
• Bioinformatics and Computational Biology

Background:

• Plastic pollution is a significant environmental and health concern.
• Discovering efficient microbial enzymes for plastic biodegradation has been challenging.
• Existing methods for enzyme discovery are often slow and inefficient.

Purpose of the Study:

• To accelerate the discovery of novel plastic-degrading enzymes.
• To develop a data-driven approach integrating multi-omics data for enzyme identification.
• To explore microbial communities adapted to plastic substrates.

Main Methods:

• Combined metagenomics, metatranscriptomics, and semi-supervised machine learning.
• Developed ranking models to prioritize candidate plastic-degrading enzymes from microbial proteomes.
• Tested models on over 10,000 microbial proteins and analyzed microbial cultures enriched on polyethylene, polyethylene terephthalate, and polyurethane.

Main Results:

• Ranking models accurately prioritized known plastic-degrading enzymes (AUC > 0.96).
• Performance was robust across different protein families.
• Identified Rhodococcus aetherivorans as a key species and several highly ranked, potentially functional enzymes expressed in enriched cultures.

Conclusions:

• Integrating multi-omics data with machine learning is a powerful strategy for enzyme discovery.
• This approach can significantly accelerate the development of biotechnological solutions for plastic pollution.
• The study highlights novel candidate enzymes with potential for plastic biodegradation.