Interpretable convolutional neural networks for sequence-based classification and discovery of plastic-degrading...

Area of Science:

• Biotechnology
• Environmental Science
• Computational Biology

Background:

• Plastic waste accumulation is a global environmental crisis.
• Biodegradation using plastic-degrading enzymes (PDEs) offers a sustainable solution.
• Current enzyme classification methods are limited by insufficient data and diversity.

Purpose of the Study:

• To develop an explainable deep learning framework (PEPIC) for classifying enzymes across nine plastic substrate types.
• To improve the accuracy and interpretability of plastic-degrading enzyme prediction.
• To accelerate the discovery of novel enzymes for plastic waste remediation.

Main Methods:

• Developed PEPIC, an interpretable convolutional neural network (CNN) model.
• Utilized a curated dataset of 181 validated PDEs and an expanded homologous dataset (~5,900 sequences).
• Benchmarked PEPIC against state-of-the-art methods, evaluating predictive performance and interpretability via amino acid contribution scores.

Main Results:

• PEPIC demonstrated statistically significant improvements in F1-score compared to existing methods.
• Amino acid contribution scores from PEPIC aligned with known catalytic residues and substrate-binding regions.
• PEPIC successfully identified a potential PET-degrading enzyme from uncurated data.

Conclusions:

• PEPIC provides a trustworthy and interpretable framework for plastic-degrading enzyme discovery.
• The AI approach accelerates the identification of enzymes crucial for plastic waste management.
• This work advances AI applications in environmental remediation and sustainable biotechnology.