iUP-BERT

iUP-BERT predicts umami peptides from polypeptide sequences using BERT-based deep learning to identify taste-active peptides.

Key Features:

• Deep learning encoding: Employs BERT (Bidirectional Encoder Representations from Transformers) to extract contextual sequence features from polypeptide sequences.
• Class imbalance handling: Uses SMOTE (synthetic minority over-sampling technique) to balance training data.
• Classification model: Utilizes support vector machine (SVM) for peptide classification and generates probabilistic scores indicating umami likelihood.
• Model validation: Evaluated using cross-validation and independent testing, demonstrating performance advantages over prior methods.

Scientific Applications:

• Umami peptide screening: Rapid identification of candidate umami peptides from sequence data to support food product and dietary supplement flavor optimization.

Methodology:

BERT-based deep representation learning for feature encoding, SMOTE for dataset balancing, SVM for classification, and evaluation via cross-validation and independent testing.