A TastePeptides-Meta system including an umami/bitter classification model Umami_YYDS, a TastePeptidesDB database and an open-source package Auto_Taste_ML

Taste peptides with umami/bitterness play a role in food attributes. However, the taste mechanisms of peptides are not fully understood, and the identification of these peptides is time-consuming. Here, we created a taste peptide database by collecting the reported taste peptide information. Eight key molecular descriptors from di/ tri-peptides were selected and obtained by modeling screening. A gradient boosting decision tree model named Umami_YYDS (89.6% accuracy) was established by data enhancement, comparison algorithm and model opti-mization. Our model showed a great prediction performance compared to other models, and its outstanding ability was verified by sensory experiments. To provide a convenient approach, we deployed a prediction website based on Umami_YYDS and uploaded the Auto_Taste_ML machine learning package. In summary, we established the system TastePeptides-Meta, containing a taste peptide database TastePeptidesDB an umami/bitter taste prediction model Umami_YYDS and an open-source machine learning package Auto_Taste_ML, which were helpful for rapid screening of umami peptides.

Automated summary

This study established a taste peptide database and a prediction model to provide a convenient method for rapid screening of umami peptides, which was validated by sensory experiments.