Identify Bitter Peptides by Using Deep Representation Learning Features.

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Bitterness in hydrolyzed proteins is caused by bitter peptides, which negatively impact palatability.
• Experimental identification of bitter peptides is time-consuming and costly, hindering product development.
• Developing efficient methods to predict bitter peptides is crucial for improving food and therapeutic products.

Purpose of the Study:

• To develop a machine learning model for accurate prediction of bitter peptides using only their sequence data.
• To leverage deep learning and advanced feature extraction techniques for enhanced bitter peptide identification.
• To provide a computational tool that aids in the development of palatable peptide-based products.

Main Methods:

• Utilized deep learning pre-trained neural network feature extraction.
• Employed sequence embedding techniques: soft symmetric alignment (SSA), unified representation (UniRep), and bidirectional long short-term memory (BiLSTM).
• Combined UniRep and BiLSTM features with a light gradient boosting machine (LGBM) for model development.

Main Results:

• The iBitter-DRLF model demonstrated significantly improved accuracy in identifying bitter peptides based on sequence data.
• Deep representation learning, particularly using UniRep and BiLSTM features, enhanced predictive capabilities.
• The developed model achieved accurate predictions solely from peptide sequence information.

Conclusions:

• The iBitter-DRLF model offers an efficient and accurate method for identifying bitter peptides.
• This computational approach can accelerate research and development for improving the palatability of peptide therapeutics and dietary supplements.
• The availability of a webserver facilitates broader application of this bitter peptide prediction tool.