A Machine Learning Method to Identify Umami Peptide Sequences by Using Multiplicative LSTM Embedded Features.

Area of Science:

• Food Science
• Biotechnology
• Computational Biology

Background:

• Umami peptides enhance food flavor and possess valuable properties for food applications.
• Current wet testing methods for identifying umami peptides are costly and time-consuming.
• Developing efficient computational methods for umami peptide identification is crucial.

Purpose of the Study:

• To develop a novel computational method for accurate umami peptide identification.
• To leverage deep learning for enhanced feature extraction from peptide sequences.
• To improve the efficiency and reduce the cost of umami peptide discovery.

Main Methods:

• Utilized a logistic regression (LR) method combined with deep learning.
• Employed a pre-trained neural network feature extraction method, Unified Representation (UniRep) based on multiplicative Long Short-Term Memory (LSTM).
• Extracted features solely from peptide sequences for iUmami-DRLF model training.

Main Results:

• Deep learning representation learning significantly improved the model's capability in identifying umami peptides.
• The iUmami-DRLF achieved high predictive precision based only on peptide sequence information.
• iUmami-DRLF demonstrated superior robustness and accuracy compared to other predictors on newly validated taste sequences.

Conclusions:

• The iUmami-DRLF model offers an efficient and accurate approach for identifying umami peptides.
• This deep learning-based method can accelerate research in umami flavor enhancement for the food industry.
• The findings support the use of sequence-based deep learning for predicting taste properties of peptides.