Identifying multi-functional bioactive peptide functions using multi-label deep learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Bioactive peptides exhibit diverse therapeutic functions, including anti-diabetic and anti-inflammatory properties.
• Machine learning is crucial for predicting peptide functions, but existing methods often focus on single functions.
• The increasing identification of multi-functional peptides necessitates advanced computational approaches.

Purpose of the Study:

• To develop a novel computational method, MLBP (Multi-Label deep learning approach for determining the multi-functionalities of Bioactive Peptides), for simultaneous prediction of multiple bioactive peptide functions.
• To address the limitations of existing single-function prediction models by enabling the identification of peptides with combined therapeutic activities.

Main Methods:

• Developed MLBP, a deep learning model utilizing peptide sequence vectors as input, bypassing traditional physiochemical features.
• Employed an embedding layer to learn dense feature vectors from sequence data.
• Integrated convolutional neural network and bidirectional gated recurrent unit layers to extract and process features for enhanced prediction.

Main Results:

• MLBP achieved an Accuracy of 0.709 and Absolute true of 0.697 on the test dataset.
• Demonstrated superior performance compared to suboptimum methods, with improvements of 5.0% and 4.7% in Accuracy and Absolute true, respectively.
• 5-fold cross-validation on the training set yielded Accuracy of 0.695 and Absolute true of 0.685.

Conclusions:

• MLBP exhibits superior prediction performance for identifying multi-functional peptides.
• The developed method offers a promising tool for discovering peptides with combined therapeutic benefits.
• MLBP is publicly available, facilitating further research in bioactive peptide function prediction.