Structure-taste correlation of L-aspartyl dipeptides using the SIMCA method

Abstract

One of the pattern recognition techniques, the SIMCA method, has been applied to structure-taste studies on L-aspartyl dipeptides (L-Asp-NH-R). The sweet and bitter taste class models of the peptides were obtained by using five structural descriptors, such as molar refractivity, and four kinds of STERIMOL parameters. The classification rates were calculated to be 87% and 81% for sweet and bitter peptides, respectively. The SIMCA method has also suggested that two factors, shape and size, of the C-terminal amino acid moiety R in the dipeptides are extremely important to model their taste qualities.

A QSTR-Based Expert System to Predict Sweetness of Molecules

“…The models presented in Table 1 from the existing literature were mainly calibrated by using small datasets and homogeneous sets of molecules, thus hampering the model generalization ability toward different types of chemicals (i.e., limited applicability domain). In addition, the majority of the studies did not perform validation of the QSTR models (Iwamura, 1980; Takahashi et al, 1982; Spillane et al, 1983; Miyashita et al, 1986b; Spillane and Sheahan, 1989, 1991). Thus, our QSTR-based expert system can be considered as a more general model for accurate prediction of sweetness of both un-evaluated and un-synthesized potential sweeteners exhibiting diverse scaffolds (i.e., a more general applicability domain).…”

“…QSTR studies regarding the prediction of the sweetness receptor-mediated taste were conducted by considering only homogeneous families of sweeteners (Iwamura, 1980; Kier, 1980; Spillane and McGlinchey, 1981; Takahashi et al, 1982, 1984; Spillane et al, 1983, 1993, 2000, 2002, 2003, 2006, 2009; Miyashita et al, 1986a,b; van der Wel et al, 1987; Okuyama et al, 1988; Spillane and Sheahan, 1989, 1991; Drew et al, 1998; Kelly et al, 2005), limiting their ability to predict the sweetness of other kinds of sweeteners. In order to generalize the predictiveness of the QSTR-based expert system, we used a dataset that covered a large chemical space of both sweet and non-sweet molecules.…”

“…Several Quantitative Structure-Taste Relationships (QSTRs) for predicting the sweetness of chemicals were proposed in the past years and are summarized in Table 1. The earlier work included compounds such as perillartine and aniline derivatives (Iwamura, 1980; van der Wel et al, 1987), sweet and bitter aldoxime derivatives (Kier, 1980), perillartine derivatives, aspartyl dipeptides, and carbosulfamates (Takahashi et al, 1982, 1984; Miyashita et al, 1986a,b; Okuyama et al, 1988), as well as sulfamate derivatives (Spillane and McGlinchey, 1981; Spillane et al, 1983, 1993, 2000, 2002, 2003, 2006, 2009; Spillane and Sheahan, 1989, 1991; Drew et al, 1998; Kelly et al, 2005). Moreover, two QSTR models to discriminate sweet, tasteless and bitter compounds have been proposed (Rojas et al, 2016a).…”

Estudio y aplicación de técnicas matemáticas de la teoría QSAR–QSPR en química analítica y química de los alimentos

“…La calidad en ajuste del modelo de Takahashi et al es comparable al obtenido en este estudio; sin embargo, la base de datos usada para su calibración está constituida únicamente por 22 compuestos, lo que lógicamente limita la realización de validación externa. En efecto, la mayoría de modelos no presentan validación externa (Iwamura, 1980;Takahashi et al, 1982;Spillane et al, 1983;Miyashita et al, 1986b;Drew et al, 1998;Spillane et al, 2002), por lo que comparar el rendimiento predictivo con respecto al modelo desarrollado no es factible.…”

“…Ese mismo año, Kier (Kier, 1980) usó 20 derivados dulces y amargos de aldoximas para desarrollar una función discriminante lineal basada en dos índices de conectividad molecular. Entre 1982y 1986, Takahashi et al (Takahashi et al, 1982Takahashi et al, 1984;Miyashita et al, 1986b) trabajaron con compuestos dulces y amargos derivados de la perillartina y del dipéptido L-aspartil para desarrollar modelos QSAR basados en máquina de aprendizaje lineal (LLA), kNN, LDA y modelado suave independiente por analogía de clases (SIMCA). Adicionalmente, Spillane et al (Spillane et al, 1983;Drew et al, 1998;Spillane et al, 2002) trabajaron en la discriminación de derivados del sulfamato mediante el uso de modelos discriminantes: gráfico, DA, LDA y QDA.…”