Parallel Convolutional Contrastive Learning Method for Enzyme Function Prediction.

Area of Science:

• Biochemistry and Bioinformatics
• Computational Biology
• Enzyme Engineering

Background:

• Enzyme function annotation is crucial for medicine and industrial biology, relying on enzyme commission (EC) numbers.
• Current enzyme function prediction tools lack the precision required for practical applications.
• Accurate prediction of enzyme functions is essential for advancing various scientific and industrial fields.

Purpose of the Study:

• To develop a novel deep learning method for improving the precision of enzyme function prediction.
• To address the challenge of class imbalance in enzyme function datasets.
• To enhance the prediction accuracy of multifunctional enzymes.

Main Methods:

• Utilized the ESM-2 protein language model for advanced protein sequence preprocessing.
• Developed a Parallel Convolutional Contrastive Learning (PCCL) framework integrating convolutional neural networks (CNNs) and contrastive learning.
• Employed a deep learning architecture with three parallel CNNs for comprehensive feature extraction.

Main Results:

• The PCCL method demonstrated improved performance over state-of-the-art enzyme function prediction techniques.
• The model showed enhanced prediction accuracy on two independent test sets.
• A significant improvement in AUC of 2.57% was observed on a smaller test set, highlighting PCCL's effectiveness.

Conclusions:

• The proposed PCCL method offers a significant advancement in enzyme function prediction accuracy.
• PCCL effectively handles class imbalance issues, leading to more robust predictions.
• This method holds promise for broader applications in medical and industrial biotechnology.