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

Conserved Sites and Recognition Mechanisms of T1R1 and T2R14 Receptors Revealed by Ensemble Docking and Molecular Descriptors and Fingerprints Combined with Machine Learning

“… a Note: index 0–9 is derived from Liang’s work, index 10–12 is derived from Richter’s work, index 13–21 is derived from Liu’s work, index 22–29 is derived from Shan’s work, index 30–38 is derived from Zhang’s work, index 39–47 is derived from Liang’s work, and index 48–53 is derived from Cui’s work …”

“…Morgan fingerprint 598, as a benzene-like ring structure, affected the water solubility of small molecules, thereby affecting the taste (umami/bitter taste) of molecules. This was consistent with the results of MD and was also supported by Umami_YYDS . Therefore, polar groups including substructures 322, 1907, 1153, 1409, 650, 509, 805, and 148 were considered to be helpful for umami (Figure A,B).…”

“…Auto_Taste_ML, which was an extension package of the sklearn and scipy package for taste peptide data processing, was selected for data dimension reduction . All feature data (DA, MD, and FP) went through 3-step screening (Figure A): (1) all data were first screened by the Kolmogorov–Smirnov test ( p ≤ 0.001) in scipy.stats.ks_2samp; (2) the second layer of screening, called scipy.stats.pearsonr, was used to perform and select 35–40 features with the highest correlation coefficient; and (3) the third layer was implemented using sklearn.feature_selection.SelectKBest to obtain the final 20 features through a forest-model-based feature importance search.…”

“…More indicators and explanations are listed in Abbreviations Used. This study chose the MolecularDe-scriptorCalculator module and produced 8 indicators selected by the Umami_YYDS model, 10 including BCUT2D_MWLOW, BCUT2-D_LOGPHI, SMR_VSA1, MinEStateIndex, VSA_EState5, VSA_ES-tate6, VSA_EState7, and MolLogP (Table S3).…”

“…Molecular descriptors are used to quantitatively describe the physical and chemical information of molecules. For example, an artificial neural network multilayer perceptron applying molecular fingerprints as input data and an umami prediction model applying the molecular descriptors as input data were constructed. To provide a good overview, the information on the published taste prediction models over the past decade was summarized (Tables S1).…”