Emily C. Lennert
Category: chemistry, statistics
Article to be reviewed:
Materazzi, S.; Gregori, A.; Ripani, L.; Apriceno, A.; Risoluti, R. Cocaine profiling: Implementation of a predictive model by ATR-FTIR coupled with chemometrics in forensic chemistry. Talanta. 2017, 166, 328-335.
Disclaimer: The opinions expressed in this review are an interpretation of the research presented in the article. These opinions are those of the summation author and do not necessarily represent the position of the University of Central Florida or of the authors of the original article.
The authors of this study examined attenuated total reflectance – Fourier transform infrared spectroscopy (ATR-FTIR) for analysis of cocaine. Cocaine is typically cut with adulterants such as caffeine or lidocaine. Characterization of cocaine and adulterant in a sample can be used to obtain a “fingerprint”, which can be used to link a sample to an original batch or source. After analysis, chemometrics, i.e. the application of statistics to scientific data, can be used to develop a predictive model. After a sample is analyzed, the predictive model can be used to predict the percent of cocaine or adulterant in the sample.
Cocaine samples were obtained and dilutions of 70, 60, 50, 40, 25, 16, and 8% were created using an adulterant (i.e. cutting agent); each cutting agent was used to create an individual series of dilutions, and cutting agents were not mixed together. Lidocaine, procaine, caffeine, phenacetin, and mannitol were selected as cutting agents. Samples were analyzed by gas chromatography – mass spectrometry (GC-MS), a standard method for drug analysis, to confirm the percent of cocaine in each sample. Samples had to be prepared for GC-MS analysis by putting the drug sample into solution. Samples were then analyzed by ATR-FTIR, no sample preparation or extractions were required. Six replicates were analyzed for each sample, and the average was used for chemometric analysis.
Principal component analysis (PCA) was performed on the ATR-FTIR data. PCA is a form of data reduction in which important data points, i.e. principal components, are identified. The goal of PCA analysis is to minimize the similar information between the samples, in an effort to maximize the differences. Once the areas that are different are observed then the authors can determine groups in which samples are really similar. Based on the samples that are in the groups, then the authors can identify what information makes the some samples more similar than others. Four principal components were identified in the data that demonstrates why certain samples grouped together; however, the principal components were not specified in the study.
Two methods were used to create predictive models: partial least squares (PLS) and principal component regression (PCR). After creation of the two models, the models were tested to determine the predictive capability of the model. Both models were determined to have a coefficient of determination, also known as r-squared value, greater than 0.9976. A coefficient of determination equal to 1 is indicative of a perfect calibration curve; the calibration curves created from the models were highly accurate. Root mean-squared error of calibration (RMSEC) was used to examine the performance of each model in calibration. An average of 2.5% error was observed in RMSEC, with both models performing similarly. This means that this model only made an error in the concentration of cocaine in the sample, 2.5% of the time. Root mean-squared error of prediction (RMSEP) was used to examine the predictive performance of each model. Both models performed similarly, and an average error of 1.74% was observed.
Upon analysis of each compound within the four models evaluated, the authors concluded that the PLS model performed better than the PCR model. The PLS model was used to calculate the percent of each analyte in samples examined by ATR-FTIR, and the calculated value was compared to the values determined by GC-MS. Values calculated based on the PLS model were found to be “very close to the expected amounts”, because it did not have a high amount of error, approximately 0.24% The authors conclude that the PLS model can be used to predict composition of cocaine samples analyzed by ATR-FTIR with low error, or the concentration and cutting agent relationship. ATR-FTIR results were comparable to GC-MS results and were obtained quickly and with no sample preparation required.
- Cocaine results obtained by ATR-FTIR were used to generate a predictive model for cocaine, which can be used to estimate the percent of cocaine and adulterants within a sample.
- The predictive model was tested and validated by several methods. A predictive model based on the partial least squares method was determined to allow for prediction with low error.
- ATR-FTIR results and the predicted results, obtained by used of the model, were compared to results obtained by GC-MS, the standard method, and were comparable.
Relevance: GC-MS is a standard technique used in many laboratories; however, the technique requires time consuming extraction and lengthy sample runs. To allow for higher throughput of casework, a faster method that can provide comparable results is required. By using a predictive model, ATR-FTIR can be used to quantify cocaine and adulterants within a sample rapidly.
- ATR-FTIR used with a predictive model provides comparable results to GC-MS, and allows for more rapid analysis with no sample preparation required for cocaine sample analysis.
- A predictive model may be a useful tool in the quantification of components within a sample with little error.