3. Results and Discussion
RMS is a collection of statistical and mathematical techniques useful for developing, improving, and optimizing processes [
22]. RMS is preferred because it minimizes the number of experiments for a specific number of factors and their levels [
23]. In general, RMS has two main types of designs—the Box–Behnken design (BBD) and the central-composite design (CCD). The BBDs differ from the CCDs in that they use fewer runs and only three levels, compared to the five in CCD. BBDs were initiated to limit the sample size as the number of parameters increased. The BBDs pair well with the green approach for the extraction of bioactive compounds, using the fewest resources possible. For this reason, the BBD was preferred in the present study.
Table 2 summarizes the obtained models of this study.
The software product Design-Expert defines the dependence of TPC on the input variables as linear (
Table 3) with statistically insignificant coefficients in front of the added members. Three factors (hydro module, duration, and temperature) influence the TPC values (
Table 4).
Other temperature-dependent studies on the topic of phenolic compounds are available in the literature [
24]. Duration is also reported as critical to phenolic content availability and possible damage and degradation [
25].
The Extractant’s effect was not statistically significant. However, many papers reveal that different extracts of the same plant result in different levels of phenolic compounds [
26]. The F-value of the model itself is 10.08 and the
p-value is <0.0001, indicating that the model is significant. The coefficient of determination is R
2 = 0.63, while adjusted R
2 = 0.56. The main idea of the “adjusted” R-squared statistic is to “penalize” the addition of terms that do not add statistical value. In this case, the only statistically insignificant term is the D-Extractant factor. Its inclusion in the TPC model caused the R-squared to decrease to a value of 0.56. In addition to the coefficient of determination, the standard error S is also applied as a measure of goodness-of-fit in the regression analysis. S is in the units of the dependent variable. In general, a higher coefficient of determination is usually associated with a smaller standard error of the regression, since a higher fit of the model to the data usually leads to smaller prediction errors. On the other hand, the standard error of the regression may decrease with a larger range of values of the independent variables or with a larger sample size, regardless of the value of the coefficient of determination. The standard error here is 5.4, meaning that the TPC data differ from the regression line by an average of 5.4. Therefore, when estimating the TPC value, an average error of less than 5.4 mg GAE/100 g fw can be expected.
Figure 1 reveals some of the most important model informational graphs. The Perturbation plot (
Figure 1a) represents the factors with the most significant influence. The hydro module’s and the duration’s graphs have the steepest slope. Since the extraction of bioactive molecules is evaluated as the first and most important step in functional ingredients to foods, pharmaceuticals, and cosmetology, among others, researchers have pointed out that an increase in the hydro module improves the diffusion rate in a solid–liquid extraction [
27].
Figure 1b presents the actual versus predicted values.
The response surface plots (
Figure 1c–e) visualize the variation in the values of two independent variables within the experimental domain while holding the other two constant.
Figure 1c reveals that the maximum TPC value can be achieved by keeping the temperature and the extractant at 70 °C and 80%, respectively. Considering the effect of hydro module and temperature on the TPC (max. 105.596 mg GAE/100 g fw),
Figure 1d predicts an optimal duration of 20 min and an 80% extractant (110.37 mg GAE/100 g fw).
Two-factor interaction terms have been added to further describe the dependence of TFC on the input variables. The results within the framework of the model appeared not statistically significant and were not visualized. An attempt was also made to define the model as linear. The results are presented in
Table 5.
It turned out that this model was also not suitable because the model’s F-value of 1.60 meant that the model was not robust to noise. There was a 20.74% chance that such a large F-value was due to noise. Finally, the 2FI model was chosen, and the results are presented in
Table 6.
It shows that the F-value is 11.68, which means that the model is significant. There is only a 0.01% chance that the F-value is due to noise. The coefficient of determination is R
2 = 0.82 with a standard error of 2.63. This confirms the fact that the higher the R-squared value, the smaller the error (compared to a TPC regression model). Thus, when estimating the TFC value, an average error of less than 2.63 mg GAE/100 g fw can be expected. The temperature does not affect the extraction of TFC (
Figure 2). Other authors also support this finding in their work, stating that the flavonoid and total phenolic contents were not influenced by temperature, time, and milling treatment [
28].
Diagnostics of the model are performed by plots of the normal distribution of residuals, residuals versus predicted values, and predicted versus actual values, presented in
Figure 2. Based on them, it can be said that the residuals are normally distributed and there are no extreme values among them.
Figure 3 presents the perturbation and contour plots. It is clear that the extractant factor has the most significant influence.
Figure 3b shows the influence of the factors hydro module and duration, while the temperature and the extractant are constants, 70 °C and 99.9%, respectively. A design point with its actual and predicted value is also presented.
Figure 3c shows the influence of the factors hydro module and extractant, while the temperature and the duration are constants, 70 °C and 40, respectively. Under these conditions, the maximum TFC value is 17.73 μg QE/100 g fw.
Figure 3d shows the influence of the factors duration and extractant, while the temperature and the hydro module are constants, 70 °C and 20, respectively.
The model describing the dependence of FRAP on the independent variables is quadratic and the results are presented in
Table 7. The F-value of the model is 23.64, which indicates the significance. There is only a 0.01 chance that an F-value this large could occur due to noise.
The coefficient of determination and the standard error in this case are, respectively, 0.92 and 13.07. Since the coefficient of determination is high, the larger value of the error is due to the significantly larger measured values (compared to the TPC and TFC models). This standard error value means that when estimating the FRAP value, a mean error of less than 13.07 μM TE/100 g fw can be expected.
Based on the diagnostics of the model shown in the plots, it can be said that the residuals are normally distributed and there are no extreme values among them (
Figure 4).
Following the perturbation and contour plots in
Figure 5, the maximum FRAP values could be obtained with a hydro module 10 for a duration of 30 min. Temperature is revealed as the least significant.
Figure 5d shows the effect of the hydro module and extractant on the FRAP while keeping duration and temperature at values of 40 min and 70 °C, respectively. Under these conditions, the maximum FRAP could be obtained with a hydro module of 10 and an extractant of 90%.
A suitable model has not been established to describe the dependence of the CUPRAC on the independent variables. A linear, 2FI, and quadratic model were tried, but all three were found to be insignificant (see
Table 8,
Table 9 and
Table 10).
The model describing the dependence of DPPH on the independent variables is quadratic (
Table 11).
The coefficient of determination R
2 and the standard error in this case are, respectively, 0.96 and 2.44. It can be seen that the coefficient of determination is high and the error value is small. This standard error value means that when estimating the DPPH value, a mean error of less than 2.44 μM TE/100 g fw can be expected. Diagnostics of the model are shown in the plots of the normal distribution of residuals, residuals versus predicted values, and predicted versus actual values, presented in
Figure 6. Based on them, it can be said that the residuals are normally distributed and there are no extreme values among them.
Figure 7 presents the perturbation and contour plots. The perturbation plot (
Figure 7a) showed that the extractant and duration have the most significant influence, while the impact of temperature had no effect.
Figure 7b revealed the effect of the hydro module and duration on the DPPH antioxidant activity of peach fruits while keeping temperature and extractant at values of 70 °C and 99.9%, respectively.
The DPPH assay has been frequently used since its method was published in 1995. Researchers are still relying on it while evaluating the antioxidant capacity of plant matrices [
29]. Under the abovementioned conditions, the maximum DPPH values (112.16 μM TE/100 g fw) could be obtained with a hydro module of 10 for a duration time of 30 min.
Figure 7c shows the effect of the hydro module and temperature on the DPPH scavenging ability while keeping duration and extractant at values of 20 and 99.9, respectively. Under these conditions, the maximum DPPH values could be obtained with a hydro module of 10 and a temperature of 60 °C.
Figure 7d shows the effect of the hydro module and extractant on the DPPH while keeping duration and temperature at values of 40 and 70 °C, respectively. Under these conditions, the maximum DPPH parameters could be obtained with a hydro module of 10 and an extractant of 90%.
Figure 7e shows the effect of temperature and extractant on the DPPH while keeping the hydro module and duration at values of 20 and 40, respectively. Under these conditions, the maximum DPPH could be obtained at 60 °C and an extractant of 90%.
Figure 7f shows the effect of duration and temperature on the DPPH while keeping the hydro module and extractant at values of 5 and 80, respectively. Under these conditions, the maximum DPPH (83.39 μM TE/100 g fw) could be obtained with a duration of 30 and a temperature of 60 °C. Other authors also present predicted values of antioxidant activity based on conditions like temperature, duration, and hydro module, stating that such results can aid in presenting the extract of choice as a functional ingredient [
30]. It has to be noted that some extracts exhibit a slower reaction with the DPPH radical, resulting in less than the actual antioxidant capacity [
31]. Thus, it is important to provide expected values under different conditions since most researchers are aiming at standardization of methods and reliability of results in different laboratories.
A quadratic model described the dependence of ABTS on the independent variables and the results are presented in
Table 12.
The coefficient of determination and the standard error in the ABTS model are 0.88 and 31.12, respectively. Since the coefficient of determination is high, it can be expected that the high error value is due to the large fluctuations (large variation range of ABTS). This standard error value means that when estimating the ABTS value, a mean error of less than 31.12 μM TE/100 g fw can be expected.
Figure 8 presents the perturbation and contour plots, where it can be seen that the hydro module, the duration, and the extractant have a quadratic influence. To the contrary, the temperature has a linear impact.
Figure 8b shows the effect of the hydro module and duration on the ABTS values while keeping temperature and extractant at 70 °C and 99.9, respectively. Under these conditions, the maximum ABTS (216.50 μM TE/100 g fw) could be obtained with a hydro module of 10 and a duration time of 30 min. Other authors state that the UAE extraction of plant matrices reveals dose-dependent ABTS values [
32].
Figure 8c shows the effect of the hydro module and temperature on the ABTS while keeping duration and extractant at values of 40 and 99.9, respectively. Under these conditions, the maximum ABTS could be obtained with a hydro module 10 and a temperature of 60 °C.
Figure 8d shows the effect of the hydro module and extractant on the ABTS while keeping duration and temperature at values of 40 and 70 °C, respectively. Under these conditions, the maximum ABTS could be obtained with a hydro module of 10 and an extractant of 90.
Figure 8e shows the effect of temperature and extractant on the ABTS while keeping the hydro module and duration at values of 5 and 20, respectively. Under these conditions, the maximum ABTS could be obtained at 60 °C and an extractant of 90.
Figure 8f shows the effect of duration and temperature on the ABTS while keeping the hydro module and extractant at values of 5 and 99.9, respectively. Under these conditions, the maximum ABTS (461.67 μM TE/100 g fw) could be obtained with a duration of 30 and a temperature of 60 °C.
A quadratic model explained the dependence of α-glucosidase on the independent variables (
Table 13). The coefficient of determination and the standard error in the Alfa-Gl model are, respectively, 0.97 and 0.0068. The extremely small error is due to both the high value of the coefficient of determination and the low measured values. This value of the standard error means that when evaluating the value of Alfa-Gl, an average error of less than 0.0068 IC 50 g/mL can be expected.
Figure 9 presents the perturbation and contour plots. The hydro module, duration, and extractant have a quadratic influence, while the temperature has a linear influence (
Figure 9a).
Figure 9b shows the effect of the hydro module and duration on the α-glucosidase while keeping temperature and extractant at 50 °C and 80%, respectively.
Under these conditions, the maximum α-glucosidase inhibition (IC
50 0.08 mg/mL) could be obtained with a hydro module of 10 for a duration time of 30 min. Other authors report twice the duration for ethanol extracts of
Azadirachta indica leaves obtained by UAE needed for the inhibition of α-glucosidase [
33].
Figure 9c shows the effect of the hydro module and temperature on the α-glucosidase inhibition potential while keeping duration and extractant at values of 20 and 80, respectively. Under these conditions, optimum results could be obtained with a hydro module of 10 and a temperature of 60 °C.
Figure 9d shows the effect of the hydro module and extractant on the α-glucosidase inhibition while keeping duration and temperature at values of 40 min and 50 °C, respectively. In this view, the optimal conditions are a hydro module of 10 and an extractant of 90.
Figure 9e shows the effect of the temperature and extractant on the α-glucosidase activity while keeping the hydro module and duration at values of 20 and 40, respectively. Under these conditions, maximal values could be obtained at 60 °C and an extractant of 90%.
Figure 9f shows the effect of the duration and temperature on the α-glucosidase inhibition potential while keeping the hydro module and extractant at values of 20 and 99.9, respectively. Under these conditions, the optimal effect (IC
50 0.13 mg/mL) could be achieved with a duration of 30 min and a temperature of 60 °C. Other authors [
34] stated that solid/solvent ratio and extraction time were key process parameters in the optimization of extraction conditions of antioxidant and α-glucosidase inhibition of weed fruits.
A quadratic model revealed the dependence of acetylcholinesterase on the independent variables (
Table 14). In this case, A, B, C, D, AB, AC, BC, BD, A2, B
2, C
2, and D2 are significant model terms. The coefficient of determination and the standard error in the AChE model are 0.98 and 0.031, respectively. The reasons for the small error are identical to those of the Alfa-Gl model—a high value of the coefficient of determination and low measured values. This standard error value means that when estimating the AChE value, a mean error of less than 0.031 IC20 g/mL can be expected.
Diagnostics of the model are shown by plots of the normal distribution of residuals, residuals versus predicted values, and predicted versus actual values, presented in
Figure 10. Based on them, it can be said that the residuals are normally distributed and there are no extreme values among them.
Figure 11 presents the perturbation and contour plots with a quadratic influence of the factors. The influence of the factors hydro module and duration is the most significant (
Figure 11a).
Figure 11b shows the effect of the hydro module and duration on the Ache inhibition potential while keeping the temperature and extractant at values of 70 °C and 80%, respectively. Under these conditions, the maximum AChE (IC
20 0.27 mg/mL) could be obtained with a hydro module of 10 for a duration time of 30 min. Proposing optimal conditions for AChE inhibition is important not only because the topic is not well-exploited yet, but also because the treatment of Alzheimer’s and Parkinson’s diseases is gradually advancing. Any information that can spare resources and time is valuable for future industrial uses. Other authors confirm that RSM can be a useful prediction tool for the plants’ AChE inhibition potential [
35].
Figure 11c shows the effect of the hydro module and temperature on the AChE while keeping the duration and extractant at values of 40 and 80, respectively. Under these conditions, the maximum Ache could be obtained with a hydro module of 10 and a temperature of 60 °C.
Figure 11d shows the effect of the hydro module and extractant on the AChE while keeping the duration and temperature at values of 20 and 50 °C, respectively. Under these conditions, the maximum Ache could be obtained with a hydro module of 10 and an extractant of 90.
Figure 11e shows the effect of the temperature and extractant on the AChE while keeping the hydro module and duration at values of 20 and 40, respectively. Under these conditions, the maximum Ache could be obtained at 60 °C and an extractant of 90%.
Figure 11f shows the effect of the duration and temperature on the AChE while keeping the hydro module and extractant at values of 5 and 80, respectively. Under these conditions, the maximum AChE could be obtained with a duration of 30 and a temperature of 60 °C. Other authors have reported the optimal conditions for UAE in terms of high AChE-inhibitory activity to be the following: methanol concentration of 85.06%, ultrasonic time of 39.1 min, and material-to-liquid ratio of 1.06:10 (g/mL) [
36].
In order for the conclusions from the dispersion analyzes to be sufficiently reliable, the following is necessary.
All samples must be drawn from normally distributed populations. To ascertain this, a normality check was performed. The Shapiro–Wilk test is a more appropriate method for small sample sizes (<50 samples), as is the case here. In this test, the null hypothesis states that the data were taken from a normally distributed population. Thus, when the
p-value > 0.05, the null hypothesis is accepted and the data are assumed to be normally distributed. The results are presented in
Table 15. The software product SPSS v.26 was used to obtain them. The reason is that the Design Expert v.13 software does not provide specific tools for performing tests for normality, homogeneity, etc., of populations such as those available in SPSS or other statistical software packages. Instead, Design Expert focuses on experimental design and analysis of experimental results. As can be seen from the last column of
Table 15, all
p-values are greater than 0.05, and therefore, the data are normally distributed. In fact, this is also evident from
Figure 2a,
Figure 4a,
Figure 6a and
Figure 10a.
The samples must have a common variance. To verify this, a homogeneity test (Test of Homogeneity of Variance) was carried out in SPSS, i.e., hypotheses of equality of population variances were tested using Levene’s Test of Equality of Error Variances. For the condition of homogeneity of variances to be met, Levene’s test should not be statistically significant, i.e.,
p-value > 0.05. The homogeneity test results obtained are presented in
Table 16. They show that the homogeneity requirement is met.
A numerical optimization is implemented, which aims to find a point that maximizes the desirability function.
Table 17 presents the constraints under which the optimization was performed. In reality, the desired effect is for values of the independent factors that are in the studied range to obtain maximum values for responses.
Table 18 shows that all goals are joined into one desirability function, which ix based on various responses and factors. The suitable optimum formulation (Hydro module of 20, Duration of 39.328 min, Temperature of 70 °C, and Extractant of 96.638%) with high desirability of 0.703 was selected. The Design Expert v.13 software returns a table of 100 possible solutions. For brevity, only the first three of them are presented here.