Identifying Sequential Residue Patterns in Bitter and Umami Peptides

Abstract

A peptide’s amino acid sequence affects its taste, but how? A rigorous structure–property connection is challenging to determine because of both the exponentially growing peptide sequence space and the scarcity of experimental measurements compared to the size of that space. By sensory methods, many peptides have been identified as tasting bitter or umami. Baselines have been determined but relate only single amino acid characteristics, in particular hydrophobicity in bitter peptides and negative charges for umami. In this work, we refine this picture by extracting sequential amino acid patterns. Our method coarse-grains the peptide sequence space to facilitate the systematic identification of common residue patterns. We identify optimal patterns for both bitter and umami peptides: one hydrophobic followed by four polar residues and two negative followed by three polar residues, respectively. We find systematic improvements compared to both random and the baselines mentioned above. Our method complements quantitative structure–activity relationship methods by leveraging sequential information to help locate taste-specific characteristics in peptides and proteins.

Classification-based machine learning approaches to predict the taste of molecules: A review

“…Subsequently, 1,613 conformation-independent Mordred features (Moriwaki et al, 2018) were subjected to feature selection by means of different approaches, which were used to calibrate diverse SVM models. The best prediction was achieved by consensus between two models Recently, Dutta proposed the identification of optimal sequential residue patterns for umami and bitter peptides (Dutta et al, 2022a). These authors used a curated database of 292 bitter and 146 umami compounds retrieved from Charoenkwan's UMP442 database (Charoenkwan et al, 2020b) and others sources.…”