Bitter peptide prediction using graph neural networks

• DOI: 10.1186/s13321-024-00909-x
• Publication Date: 2024-10-07T21:01:46Z
• Authors: Prashant Srivastava; Alexandra Steuer; Francesco Ferri; Alessandro Nicoli; Kristian Schultz; Saptarshi Bej; Antonella Di Pizio; Olaf Wolkenhauer
• Keywords: Tetrapeptide; Bitter Taste; Tripeptide

Bitter taste is an unpleasant modality that affects food consumption. peptides are generated during enzymatic processes produce functional, bioactive protein hydrolysates or the aging process of fermented products such as cheese, soybean protein, and wine. Understanding underlying peptide sequences responsible for bitter can pave way more efficient identification these peptides. This paper presents BitterPep-GCN, a feature-agnostic graph convolution network prediction. The graph-based model learns embedding amino acids in uses mixed pooling classification. BitterPep-GCN was benchmarked using BTP640, publicly available dataset. latent embeddings by trained were used to analyze activity sequence motifs Particularly, we calculated individual dipeptide, tripeptide, tetrapeptide present Our analyses pinpoint specific acids, F, G, P, R, well motifs, notably tripeptide containing FF, key signatures work not only provides new predictor various but also gives hint into molecular basis bitterness.Scientific ContributionOur first application Graph Neural Networks prediction taste. best-developed model,