Hyun-Ock Kim | PubMed Health

PubMed2006

Quantitative structure-activity relationship study of bitter peptides.

A database consisting of 224 di- to tetradecapeptides and five amino acids was compiled to study quantitative structure-activity relationships of bitter peptides. Partial least-squares regression-1 analysis was conducted using the amino acid three z-scores and/or three parameters (total hydrophobicity, residue number, and log mass values) as X-variables and bitterness values (log 1/T where T is the bitterness threshold) as Y-variables. Using the three parameters only, significant models (p < 0.001) were obtained describing the entire data set as well as data subsets, except that comprised only of octa- to tetradecapeptides. For data sets comprising different peptide lengths, the models were improved by including the three z-scores at the N-terminal and C-terminal positions. Correlation coefficients for bitterness prediction of 48 dipeptides and 12 pentapeptides were 0.75 (RMSEP = 0.53) and 0.90 (RMSEP = 0.48), respectively. Bulky hydrophobic amino acids at the C terminus and bulky basic amino acids at the N terminus were highly correlated to bitterness.

Amino Acid Sequence↗

PubMed2006

Application of Fourier transform Raman spectroscopy for prediction of bitterness of peptides.

The potential application of Fourier transform (FT) Raman spectroscopy to predict the bitterness of peptides was investigated. FT-Raman spectra were measured for the amino acid Phe and 9 synthetic di-, tri-, and tetra peptides composed of Phe, Gly, and Pro. Partial least squares regression (PLS)-1 analysis was applied to correlate the FT-Raman spectra with bitterness intensity values (R(caf) and log 1/T) reported in the literature. Using full cross-validation, Model 1 based on the single spectral data set for the nine peptides yielded a high correlation coefficient for calibration (R = 0.99), but a low correlation coefficient for prediction (R = 0.56). Two models were constructed using the data sets including replicate spectra for the calibrations and were validated using full cross-validation. Using leave-one-sample-set-out calibrations, Model 2, which was developed with the data for the peptides as well as Phe, yielded a low correlation coefficient (R = 0.533) for the prediction of the bitterness, while Model 3 developed with only the peptide data provided better correlation coefficients (R = 0.807 and 0.724 for R(caf) and log 1/T values, respectively). The correlation coefficients for prediction were 0.975 (R(caf) values) and 0.874 (log 1/T values) for Model 4, which was developed using subtracted spectral data (spectra of peptides with higher R(caf) values minus spectra of peptides with lower R(caf) values). Examination of the PLS regression coefficients at wavenumbers most highly correlated with bitterness revealed the importance of hydrophobicity and peptide length on bitterness. This study indicates the potential of FT-Raman spectroscopy as a useful tool for predicting bitterness of peptides and amino acids.

Amino Acids↗