Matrix-Matched Calibration for the Quantitative Analysis of Pesticides in Pepper and Wheat Flour: Selection of the Best Calibration Model

Abstract

An automated package for calculating the best calibration model for matrix-matched calibration in food pesticide analysis has been developed in this study. The algorithm development in the package is based on three requirements for routine food pesticide analysis: a good working range fitness for samples with high maximum residue limits (MRLs), detection capability for pesticide analysis with MRLs close to the limit of quantitation, and a simple working range problem detection model. The requirements are combined in a simple scoring system above 100. The package has been tested in the analysis of pesticides of pepper and wheat flour. The results show that the package can be used for different pesticides quickly and visually, and also allows evaluation of matrix effects between different matrix calibrations. For the pesticides tested with the package, the weighted linear calibration gave the best score over the simple linear calibration and second-order calibration.

1. Introduction

Pesticides are widely used in agriculture to protect livestock and crops, with more than a million tonnes used annually worldwide to control a wide range of pests [1]. Around 333 thousand tonnes of pesticides were used in 2021, with more than 400 new pesticides on the market [2]. In this sense, it is important to monitor pesticide residues in order to ensure the safety and quality of food due to the widespread use of pesticides on crops. To prevent health issues related to excessive pesticide use, maximum residue limits (MRLs) have been established by regulations worldwide [3]. For example, in the European Union (EU), the European Commission’s consolidated version of Regulation 396/2005 specifies MRLs for various food products [4,5].

The development of appropriate methods to assess food safety, according to the requirements of international quality standards, is one of the main objectives of pesticide analysis [6]. To determine the maximum number of pesticides in a cost-effective and efficient manner, multi-residue methods (MRMs) are necessary. Until now, pesticides have usually been determined by classical analytical methods such as gas chromatography (GC) and liquid chromatography (LC) coupled with mass spectrometry (MS), and the most widely used pesticide residue extraction method in routine laboratories today is the method with the acronym “Quick, Easy, Cheap, Effective, Rugged, and Safe” (QuEChERS). The QuEChERS method involves two steps, the first being an initial extraction with different salt formulations to drive the separation of the organic extraction solvent and water. The second step involves an aliquot of the organic phase undergoing a cleanup process through dispersive solid-phase extraction (d-SPE) using different sorbents to remove the matrix components before chromatographic analysis [7].

The QuEChERS extraction method has undergone various modifications to accommodate a wide range of products. For example, water may be added for spices, flour, and other dehydrated matrices, or a buffer method may be used when analysing pH-sensitive pesticides, as suggested by EN 15662 [8].

The clean-up approaches aim to reduce the matrix effect (ME) in the variability of foods that can be analysed, such as foods with chlorophyll or carotenoids and other natural colours (e.g., spices), fat or lipid content (e.g., nuts), essential oils and flavonoids (e.g., herbs), etc. [9]. The ME leads to analytical problems that affect the accuracy of the results, which can be observed in the recovery of pesticides, leading to either a reduction or an increase in the acceptable value for validation from 70% to 120% [6,10].

One of the most common approaches to reducing the ME is the matrix-matched calibration (MMC) method [11]. Some of its advantages include the wide range of matrices that can be used [9,10,12,13], its use in GC and LC, and its use in conjunction with other applications such as the use of internal standards [14]. The use of MMC is widely described for calibration purposes [15,16] and is used in various fields of instrumental analysis [17,18,19], particularly in MS. MMC is utilized to account for the effects on the ionization efficiency of the compounds being studied, which may be influenced by the co-presence of different organic compounds alongside the analytes. This co-presence can lead to suppression or enhancement of the analyte signal, impacting the reproducibility and accuracy of the results [20].

In the analysis of pesticides, using MMC is a recommended option in various guides and standards for pesticides [14]. MMC has been widely studied in different and complex samples analysed by GC or LC [21,22,23,24] and is considered suitable for routine laboratory use. It ensures the precision, recovery, and uncertainty requirements outlined in the SANTE 11312/2021 [6] guideline for pesticide analysis. However, although there are many studies on the influence of the matrix on pesticide analysis and quantification errors [25], there is no information on the selection of appropriate calibration functions. The most appropriate calibration functions, based on those described in the literature [6,14], should be selected using the simplest model among linear, weighted linear, or second-order.

While the influence of this factor is not expected to significantly impact the overall uncertainty of pesticide analysis (since precision and recovery are the most crucial contributing factors [26]) it is still important to select a calibration model that minimizes this influence. Alternatively, being able to quickly visualize which factors affect the choice of one model over another in routine analysis is also necessary. Moreover, for pesticide analysis, the calibration model must enable the most accurate quantification of concentrations near the lowest calibration point for pesticides with very low maximum residue limits (MRLs) like carbofuran (MRL: 0.002 mg/kg in sweet pepper), while also being able to quantify other pesticides with much higher MRLs at the upper end of the calibration range (e.g., fluopyram with an MRL of 2 mg/kg in sweet pepper). In this scenario, the algorithm used becomes particularly important.

Further, commercial chromatography software offers numerous numerical and graphical tools; however, it lacks the capability to compare different calibration models for all pesticides at the same time. Additionally, it is unable to show how the use of a model impacts the calibration models for different pesticides and its influence on parameters that affect result quantification, such as detection capability and goodness of fit (GOF), as per routine validation based on the SANTE [6] requirements.

Therefore, the aim of this work was to create an R package, which is publicly available on GitHub (https://github.com/chemaveba/ChemACal, accessed on 20 May 2024), to evaluate and select the best calibration model using various tools and a scoring system based on SANTE and ISO requirements for the analysis of pesticides. Its aim is to streamline the process for routine laboratories in selecting the most accurate model for different MME calibrations. This includes reducing the time required for model selection and ensuring minimal errors in routine analysis through numerical and graphical analysis. To assess the package’s performance, 23 pesticides were analysed using LC and GC with pepper as the matrix line. Additionally, the package was utilized to identify discrepancies in MME between pepper and wheat flour for GC-analysed pesticides.

2. Materials and Methods

2.1. R Package

The R package was developed using the R-Studio IDE Desktop 2023.09.0 tool, and R 4.2.2 (59 Temple Place, Suite 330, Boston, MA 02111-1307, USA)was used to analyse the data and generate the graphs [27]. This software is free and can be used in both desktop and cloud versions. R-Studio allows the creation of scripts and use of loops with code written in R. The R package ChemACal (chemical analytical calibration) can be used to automate on loop all pesticides analysed by GC-MS/MS and LC-MS/MS simultaneously. An example of the loop can be provided upon request to the Authors.

2.2. ChemACal Package

The proposed workflow for analysing the calibration data involves three sequential steps (Figure 1). In step 1, data are exported from the commercial software connected to the chromatography system. In step 2, the R package requires two different datasets containing the necessary information to evaluate the calibration. The data must be imported in CSV, XLSX, or other compatible formats. The first dataset should include various columns with the names of all the pesticides, while the second dataset should contain all the calibration data, with two columns for each pesticide named in the first dataset. The first column of the second dataset should contain the concentration information, and the second column should contain the instrumental response (area) for each concentration. As a precaution, for each pair of concentration and instrumental response values, it is important to verify that the concentration data values match the response. If any discrepancies are found, the analysis should move on to the next pesticide.

Step 3, described in Section 2.2.1 and Section 2.2.2, is the evaluation and scoring algorithm used in the package.

2.2.1. Evaluation of Linearity and Calibration Model

There are several alternatives to evaluate linearity based on graphical outputs (residual plot, linearity plot), statistical tests (ANOVA, ANOVA-LOF, Mandel, significance of the quadratic term), or numerical parameters (R2, QC, or quality coefficient) in the instrumental analysis [15,28,29,30,31,32] since their use is appropriate for the evaluation of the calibration model, taking into account their advantages and disadvantages.

The algorithm that was used focused on evaluating, on the one hand, the GOF in all the range of calibration and, on the other hand, the capability of detection (COD) of each calibration model. According to Section 6.2.1 in the UNE-CEN/TS 17061 [14], the simplest acceptable calibration, the linear, the weighted linear, and the second-order models were calculated.

Several assumptions used in routine pesticide analysis were made, such as that the calibration standard is injected only once, and for this, the simplest weight (w) used was 1/x, where x is the concentration at each point of the calibration.

Equation (1) shows the methodology used for the evaluation of the GOF on the basis of the residual standard deviation (S_yrel) according to Mandel [33].

The calculation of the GOF can be found in the standards ISO 8466-1 [33] and ISO 8466-2 [34] but expressed as relative residuals (S_y) [30]. On the other hand, Equations (2)–(5) show the algorithm for evaluating the detection capability based on ISO 11843-2 [35] and ISO 8466-2 [34].

$${\mathrm{S}}_{\mathrm{y}\mathrm{r}\mathrm{e}\mathrm{l}}=100\times \sqrt{\frac{\sum _{\mathrm{i}=1}^{\mathrm{N}}{[\frac{{\mathrm{y}}_{\mathrm{i}}-{\widehat{\mathrm{y}}}_{\mathrm{i}}}{\overline{y}}]}^2}{\mathrm{N}-\mathrm{p}}}$$

(1)

where N is the number of calibration points; p is the point for calculating the degree of freedom; the 2 is for linear cases, while 3 is used for second-order calibration; the y_i represents the experimental response; ${\overline{\mathrm{y}}}_{\mathrm{i}}$ is the response applied the model of calibration; and $\overline{\mathrm{y}}$ is the average response for transformation to relative S_y.

Equation (2) shows the COD of each calibration model in the linear case.

$$\mathrm{C}\mathrm{O}\mathrm{D}=\mathsf{\delta }\times \frac{S_y}{b}\times \sqrt{A+\frac{1}{N}+\frac{{\overline{x}}^2}{S_{xx}}}$$

(2)

where δ is the value of the non-centrality parameter with N-2 degrees of freedom and a relative error of 5%; b is the slope for the calibration model; $\overline{x}$ is the mean concentration for the calibration points expressed by Equation (3); and S_xx is the sum of the squared deviations of the calibration points expressed by Equation (4).

Equation (3) shows the sum of the squared deviations of the calibration points.

$${\mathrm{S}}_{\mathrm{x}\mathrm{x}}={\sum }_{\mathrm{i}=1}^{\mathrm{N}}{w}_i{(x_i-\overline{x})}^2$$

(3)

Equation (4) shows the mean concentration for the calibration points.

$$\overline{x}=\frac{\sum _{\mathrm{i}=1}^{\mathrm{N}}{w}_i\times x_i}{F}$$

(4)

In the case of linear calibration without weight, the value of w_i is 1, and F is the number of calibration points. If the weight 1/x is used, w_i is 1/x_i and the value of F is $\sum _{\mathrm{i}=1}^{\mathrm{N}}{w}_i$.

Equation (5) shows the COD of each calibration model in the second-order case.

$$\mathrm{C}\mathrm{O}\mathrm{D}=\mathsf{\delta }\times \frac{S_y}{b_s}\times \sqrt{A+\frac{1}{N}+\frac{{\left(\widehat{x}{-\overline{x}}^2\right)}^2\times Q_{x4}+{({\widehat{x}}^2-\frac{\sum _{\mathrm{i}=1}^{\mathrm{N}}{x}_i^2}{N})}^2\times Q_{xx}-2\times (\widehat{x}-\overline{x})\times ({\widehat{x}}^2-\frac{\sum _{\mathrm{i}=1}^{\mathrm{N}}{x}_i^2}{N})\times Q_{x3}}{Q_{x4}\times Q_{xx}{-(Q_{x3})}^2}}$$

(5)

where b_s is the second-order slope (in this case it is b + cx), and the terms Q_xx, Q_x3, and Q_x4 are intermediate values as defined in Section 4 of ISO 8466-2 [34].

The term “A” of Equations (2) and (5) is 1/r, where r is the number of replicates for each point of the calibration. In our case, since no replications were carried out, the value of 1/r was estimated as “1”. It should be noted that although the use of replicates would improve the COD, this was not taken into account because it is impractical or impossible to carry out in routine analysis of pesticides in food samples, where many pesticides are analysed simultaneously at different calibration points.

2.2.2. Selection of the Model of Calibration

To assist in the selection of the calibration model, a difference DS² in the sums of squared deviations is calculated from the residual standard deviation according to Equation (6).

$${DS}^2=\left(\mathrm{N}-2\right)\times S_{yl}^2-\left(\mathrm{N}-3\right)\times S_{ys}^2$$

(6)

where S_yl is the residual standard deviation in the linear case (linear and linear-weighted); and S_ys is the residual standard deviation in the second-order calibration.

By applying an F-test [33] to the value of DS² and S_ys, the calibration function can be estimated: it is linear in the working range examined. If it is concluded that it is not linear, a check shall be made whether there is a maximum or minimum within the working range by application of Equation (7).

$$x^{*}=-\frac{b}{2c}$$

(7)

where x* must be between the first and last calibration point. The terms b and c refer to the second-order calibration coefficient (y = a + bx + cx²).

2.2.3. Algorithm Used to Select the Best Model of Calibration

The algorithm used to score the best calibration model is based on a simple scoring system described in Figure 2.

Block 1: forms source data from software;

Block 2: sets initial conditions of the concentrations (x) and instrumental response (y);

Block 3 to 4: calculate the values of parameters GOF (Block 6) and COD (Block 7) according to Equations (1)–(5) described in Section 2.2.1. The system awards a maximum score of 30 to the lowest value obtained from GOF and COD for each of the models under consideration. For instance, if the lowest GOF value is 10 for the linear model, the system assigns it a score of 30. From this value, the score is calculated using a simple inverse rule of three. Therefore, if the GOF values are 13 and 17 for the linear-weighted and second-order models, the final scores will be 23 and 18, respectively;

Block 5: The other 40 points (Block 8) are allocated to the model based on the Mandel test as defined in ISO 8466-1 [33], giving priority to linear calibrations over second-order calibrations [as defined by UNE-CEN/TS-17061 [14]];

Block 9: calculate the sum of the points obtained in blocks 6 to 8 to obtain the final score for each calibrated model.

2.2.4. Calculation of the Matrix Effect

Equation (8) shows the matrix effect between the MMC used. The ME was calculated from slopes of calibration curves of the different matrices used [36] and evaluated according to the established range of low (<20%), moderate (20–50%), and high (>50%) [21].

$$ME\left(\%\right)=100\times (\frac{Slopematrix2}{Slopematrix1}-1)$$

(8)

where slope matrix 1 is the slope in the calibration model selected for the MMC used as reference (in this case of pepper) and slope matrix 2 is the slope in the same calibration model used to evaluate the ME (in this case of wheat flour).

2.3. Pesticide Analysis

The QuEChERS extraction method, based on the E1 extraction of UNE [37], involved using 10 g of pepper and 5 g of wheat flour in a 50 mL polypropylene tube without adding water. Then, 10 mL of acetonitrile was added and the mixture was shaken for 1 min. After shaking, the salt extraction kit was added, and the mixture was centrifuged at 3000 rpm for 5 min. A clean-up step using C2 EN 15662 of UNE [37] and PSA was performed on the organic aliquot obtained and, finally, it was centrifuged again at 3000 rpm for 5 min to obtain the final aliquot for the GC-MS/MS and LC-MS/MS analyses.

A total of 10 pesticides were analysed by GC-MS/MS in pepper and wheat flour matrices using an Agilent 7890 (Santa Clara, CA, USA) GC system coupled to a 7000A quadrupole tandem mass spectrometer. The chromatographic separation was performed on a RESTEK RTx-5MS column (30 m, 0.25 mm, i.d., 0.5 µm). Helium was used as the carrier gas at a constant flow rate of 1.2 mL/min, while argon was used as the collision gas. The oven temperature was adjusted as follows: the initial temperature was set at 70 °C for 2 min, then increased to 150 °C at a rate of 25 °C min⁻¹, to 200 °C at 3 °C min⁻¹ with a hold time of 1 min, and finally to 280 °C at 10 °C min⁻¹ with a hold time of 10 min. The total run time was 45 min. The temperatures of the transfer line and ion source were set at 280 °C and 250 °C, respectively. The mass spectrometer was operated in multiple reaction monitoring (MRM) mode with three mass transitions. The calibration range used for pepper and wheat flour was between 3 and 100 ng/mL.

A total of 13 pesticides were analysed by LC-MS/MS in pepper matrix using an Agilent liquid chromatography system (Santa Clara, CA, USA) coupled to a 6470 triple quadrupole tandem mass spectrometer. The chromatographic separation was performed on a Poroshell C18 column (150 mm, 2.1 mm, i.d., 2.7 μm) (Agilent, USA) with a flow rate of 0.1 mL at 40 °C. The elution solvents used were water 5 mM ammonium formate with 0.01% formic acid (A) and methanol 5 mM ammonium formate with 0.01% formic acid (B). The gradient elution was performed as follows: 40% solvent B for 0–5 min, changing to 60% solvent B for 6–12 min, and finishing with 100% solvent B for 17–20 min. Pesticides were analysed using programmed MRM in positive and negative modes simultaneously. Ion source parameters included optimised drying gas temperature, drying gas flow rate, nebuliser pressure, sheath gas temperature and flow rate, capillary voltage, nozzle voltage, and high and low radiofrequency voltage. The calibration range for pepper samples was between 1 and 50 ng/mL.

MRM transitions and retention times for the pesticides tested are given in Table S1.

3. Results and Discussion

3.1. Evaluation of the Calibration Data

The GC-MS/MS and LC-MS/MS calibration data exported and processed by the doEvalCal function are shown in

Table 1. Results applying doEvalCal function to matrix-matched calibration by GC-MS/MS.

Compound	Model	COD	GOF	Score
Chlorpropham	Linear	2.27	1.5	43
	Linear-Weighted	0.53	1.5	86
	Second-order	1.43	0.8	41
Fenpropathrin	Linear	3.13	2.1	33
	Linear-Weighted	0.645	2.3	77
	Second-order	0.805	0.5	54
Endimethalin	Linear	4.6	3.1	9
	Linear-Weighted	0.964	3.5	65
	Second-order	0.719	0.4	80
Cypermethrin	Linear	7.83	4.9	33
	Linear-Weighted	1.7	5.5	75
	Second-order	1.97	1	56
Cyproconazole	Linear	2.98	2	64
	Linear-Weighted	0.766	2	66
	Second-order	1.87	1.1	42
Cyprodinil	Linear	1.7	1.1	84
	Linear-Weighted	0.801	1.2	78
	Second-order	2.08	1.2	40
Dimethomorph	Linear	3.12	2.2	75
	Linear-Weighted	1.03	2.2	75
	Second-order	2.98	1.8	40
Fludioxonil	Linear	0.996	0.6	92
	Linear-Weighted	0.714	0.7	76
	Second-order	1.32	0.8	38
Iprodione	Linear	2.63	1.7	85
	Linear-Weighted	1.28	1.7	80
	Second-order	3.15	1.8	40
Pyriproxyfen	Linear	5.22	3.3	41
	Linear-Weighted	1.25	3.6	82
	Second-order	2.81	1.5	43

COD: capability of detection in ng/mL; GOF: goodness of fit in percentage; Score: scoring after applying the evaluation in Figure 2.

and

Table 2. Results applying doEvalCal function to matrix-matched calibration by LC-MS/MS.

Compound	Model	COD	GOF	Score
Mandipropamid	Linear	2.31	4.2	61
	Linear-Weighted	0.835	4.2	100
	Second-order	2.78	4.3	38
Spirotetramat (sum)	Linear	1.35	2.5	82
	Linear-Weighted	0.563	2.8	77
	Second-order	1.56	2.5	41
Spirotetramat enol	Linear	2.94	5.1	82
	Linear-Weighted	1.2	5.4	78
	Second-order	3.59	5.8	36
Acetamiprid	Linear	2.79	5	53
	Linear-Weighted	0.753	5.3	93
	Second-order	2.8	4.1	38
Boscalid	Linear	1.08	2	82
	Linear-Weighted	0.425	2.3	76
	Second-order	1.31	2.2	37
Chlorantraniliprole	Linear	2.8	5.1	49
	Linear-Weighted	0.973	5.4	88
	Second-order	1.79	3.3	46
Fenhexamid	Linear	3.41	6	15
	Linear-Weighted	0.926	7.2	73
	Second-order	0.824	1.5	80
Metaflumizone	Linear	3.77	6.6	53
	Linear-Weighted	1.16	7.3	91
	Second-order	2.88	5.2	42
Pyraclostrobin	Linear	0.653	1.2	88
	Linear-Weighted	0.382	1.5	74
	Second-order	0.836	1.4	40
Spinosad	Linear	2.55	4.4	43
	Linear-Weighted	0.732	5	83
	Second-order	1.17	2.1	49
Spirodiclofen	Linear	3.7	6.4	60
	Linear-Weighted	1.28	6.6	99
	Second-order	4.81	7.2	35
Spiromesifen	Linear	3.16	5.9	62
	Linear-Weighted	1.22	6.8	96
	Second-order	3.79	5.9	40
Thiacloprid	Linear	5.77	10.3	50
	Linear-Weighted	1.46	10.9	91
	Second-order	3.91	7.7	41

COD: capability of detection in ng/mL; GOF: goodness of fit in percentage; Score: scoring after applying the evaluation in Figure 2.

. Table 2 shows only the information calculated by Equations (1)–(6) and the score obtained but other information such as the regression equation, the coefficient of determination, and the minimum response for the first calibration point are shown for comparison with the commercial software used by the chromatograph.

Table 1 and Table 2 show that the model with the best score in most of the pesticides is in the linear-weighted calibration mainly due to a minimum COD and a good GOF (62% of pesticides studied), which ensures that values can be quantified close to the limit of quantification and that the fitness is adequate over the entire working range studied.

On the other hand, the GOF results do not show that the use of a second-order calibration (15% of pesticides) is always an improvement compared to a linear calibration (23% of pesticides). Clear examples of such cases are cyprodinil, dimetomorph in GC-MS/MS, or mandipropamid and spirotetramat in LC-MS/MS, where there are cases of second-order values that are higher than the linear case. For example, spriodiclofen has a second-order GOF value of 7.2 versus 6.6 in the linear-weighted case. The same situation is repeated for the calculation of COD, where the linear-weighted calibration generally gives a better COD than the linear or second-order calibration, e.g., boscalid, acetamiprid, or pyraclostrobin. The combination of the two factors, GOF and COD, together with minimal model error, gives the best results in the linear-weighted calibration compared to the second-order calibration.

The values of GOF are similar in the case of linear calibration versus the linear-weighted calibration, e.g., spiromesifen has a GOF value of 5.9 in the linear case versus 6.8 in the linear-weighted case. However, for COD, the linear-weighted calibration gives better results than the linear calibration in all cases (1.22 versus 3.66 ng/mL).

The results confirm the statement in the UNE-CEN/TS-17061 [14] standard that the use of linear calibration models is preferable to more complex models such as second-order models. In previous studies on pesticide analysis, researchers have discussed the choice between linear-weighted calibration and linear calibration [38]. Similarly, the statistical evaluations conducted on calibration models, such as the calculation of COD or the type of model used, have also been considered [39,40]. However, none of these studies have combined all these evaluations into a single methodology.

3.2. Graphical Visualization

The package includes two graphical tools. The first tool enables filtering by calibration type and COD limit. For example, if the pesticide being analysed must have a limit of quantification of at least 0.005 mg/kg, then the calibration data must have a COD limit of 0.003 mg/L. The doGrafXcrit function provides a quick visualization of the number of compounds that meet this condition, as demonstrated in Figure 3.

Figure 3 also shows that the linear calibration does not meet the COD limit (x-axis) of 0.003 µg/mL for 57% of the compounds (y-axis). So, a priori, it would not be the best calibration to use for the pesticides analysed. To be able to visualise which is the best calibration by comparing the different calibrations, a new graphical tool is needed. The visualisation of all the information obtained by the doEvalCal function is realised by the doGrafBcal function. This function is used to obtain data from GOF (Equations (1)–(3)), COD (Equations (4)–(6)), the calculated score, and the instrumental response of the first point of calibration. The graphical results are grouped by type of calibration and with a radius in the bubble chart depending on the value of COD.

In Figure 4, three colours represent different types of calibration: purple for linear calibration, green for weighted calibration, and yellow for second-order calibration. This graph enables an interactive display of all the information and compound names. For example, thiacloprid’s information can be conveniently displayed alongside all the other data points.

As can be seen in Figure 4, the linear-weighted calibration (green bubbles) gives the best results as the bubbles have a smaller radius or size (better COD), a higher x value or SCHEMA score and, generally, lower GOF values (y-axis). The worst calibration is the linear calibration for the worst COD and, therefore, a low value in the SCHEMA score. The two bubbles with the two worst GOF values (values above 9 on the y-axis) correspond to the compound thiacloprid. The thiacloprid in the linear-weighted calibration has a higher GOF value (10.9) but with a COD of 0.0015 µg/mL, it is sufficient to meet the limit of 0.003 µg/mL, which would not be met in the linear calibration with a value of 0.0058 µg/mL.

Figure 5 shows the graphical option doGraphME. This option allows the visualisation of the pesticides classified according to their matrix effect (low, medium, and high). For the pesticides analysed with different MME calibrations (pepper and wheat flour), the results of Figure 5 show that only cyprodinil has a high matrix effect (>100) and the rest of the pesticides have a low matrix effect.

4. Conclusions

The results of this study showed that the new R package could be used to select the best calibration in a pesticide analysis where many pesticides could be analysed.

Using the package, the best model calibration obtained was the linear-weighted calibration, where a better evaluation was based on good GOF and COD values compared to the other models evaluated—the classical linear calibration and the second-order calibration.

This package was also found to be suitable for testing matrix effects against different types of matrix calibration.

All information can be quickly visualised through datasets or various graphical tools implemented in the package.