Unraveling surface-enhanced Raman spectroscopy results through chemometrics and machine learning: principles, progress, and trends.

Surface-enhanced Raman spectroscopy (SERS) has gained increasing attention because it provides rich chemical information and high sensitivity, being applicable in many scientific fields including medical diagnosis, forensic analysis, food control, and microbiology. Although SERS is often limited by the lack of selectivity in the analysis of samples with complex matrices, the use of multivariate statistics and mathematical tools has been demonstrated to be an efficient strategy to circumvent this issue. Importantly, since the rapid development of artificial intelligence has been promoting the implementation of a wide variety of advanced multivariate methods in SERS, a discussion about the extent of their synergy and possible standardization becomes necessary. This critical review comprises the principles, advantages, and limitations of coupling SERS with chemometrics and machine learning for both qualitative and quantitative analytical applications. Recent advances and trends in combining SERS with uncommonly used but powerful data analysis tools are also discussed. Finally, a section on benchmarking and tips for selecting the suitable chemometric/machine learning method is included. We believe this will help to move SERS from an alternative detection strategy to a general analytical technique for real-life applications.

A fast and effective approach for the discrimination of garlic origin using wooden-tip electrospray ionization mass spectrometry and multivariate classification.

This paper presents the combination of wooden-tip electrospray ionization mass spectrometry (WTESI-MS) and multivariate pattern recognition methods (principal component analysis, PCA and partial least squares discriminant analysis, PLS-DA) for the rapid and reliable discrimination, via chemical fingerprints, of garlic origin. A total of 312 garlic samples grown in different countries (Brazil, China, Argentina, Spain, and Chile) were studied. The methodology was based on a direct sampling approach, which relies on loading the sample by penetrating the garlic cloves with a pre-wetted wooden tip, followed by direct prompt analysis by WTESI-MS. Thus, no sample preparation is needed, which prevents the degradation of important metabolites and increases the analytical throughput. Parameters that affects the WTESI were optimized and the best performance in terms of signal stability and intensity was achieved using the positive ion mode. Most of the ions in WTESI mass spectra were assigned to amino acids, sugars, organosulfur compounds, and lipids. The discriminative model showed good performance (accuracy rates between 81.9% and 98.6%) and enabled identifying diagnostic ions for garlic samples from different origins. The differentiation and classification of garlic origin is of major importance as this food flavoring product is widely consumed, with worldwide trade representing billions of dollars every year, and is very often the subject of fraud.

Amino Acid Biosignature in Plasma among Ischemic Stroke Subtypes.

Ischemic stroke is a neurovascular disorder caused by reduced or blockage of blood flow to the brain, which may permanently affect motor and cognitive abilities. The diagnostic of stroke is performed using imaging technologies, clinical evaluation, and neuropsychological protocols, but no blood test is available yet. In this work, we analyzed amino acid concentrations in blood plasma from poststroke patients in order to identify differences that could characterize the stroke etiology. Plasma concentrations of sixteen amino acids from patients with chronic ischemic stroke (n = 73) and the control group (n = 16) were determined using gas chromatography coupled to mass spectrometry (GC-MS). The concentration data was processed by Partial Least Squares-Discriminant Analysis (PLS-DA) to classify patients with stroke and control. The amino acid analysis generated a first model able to discriminate ischemic stroke patients from control group. Proline was the most important amino acid for classification of the stroke samples in PLS-DA, followed by lysine, phenylalanine, leucine, and glycine, and while higher levels of methionine and alanine were mostly related to the control samples. The second model was able to discriminate the stroke subtypes like atherothrombotic etiology from cardioembolic and lacunar etiologies, with lysine, leucine, and cysteine plasmatic concentrations being the most important metabolites. Our results suggest an amino acid biosignature for patients with chronic stroke in plasma samples, which can be helpful in diagnosis, prognosis, and therapeutics of these patients.

Screening method for rapid classification of psychoactive substances in illicit tablets using mid infrared spectroscopy and PLS-DA.

Several new psychoactive substances (NPS) have reached the illegal drug market in recent years, and ecstasy-like tablets are one of the forms affected by this change. Cathinones and tryptamines have increasingly been found in ecstasy-like seized samples as well as other amphetamine type stimulants. A presumptive method for identifying different drugs in seized ecstasy tablets (n=92) using ATR-FTIR (attenuated total reflectance - Fourier transform infrared spectroscopy) and PLS-DA (partial least squares discriminant analysis) was developed. A hierarchical strategy of sequential modeling was performed with PLS-DA. The main model discriminated four classes: 5-MeO-MIPT, methylenedioxyamphetamines (MDMA and MDA), methamphetamine, and cathinones. Two submodels were built to identify drugs present in MDs and cathinones classes. Models were validated through the estimate of figures of merit. The average reliability rate (RLR) of the main model was 96.8% and accordance (ACC) was 100%. For the submodels, RLR and ACC were 100%. The reliability of the models was corroborated through their spectral interpretation. Thus, spectral assignments were performed by associating informative vectors of each specific modeled class to the respective drugs. The developed method is simple, fast, and can be applied to the forensic laboratory routine, leading to objective results reports useful for forensic scientists and law enforcement.

Detection and characterisation of frauds in bovine meat in natura by non-meat ingredient additions using data fusion of chemical parameters and ATR-FTIR spectroscopy.

Concerns about meat authenticity are increasing recently, due to great fraud scandals. This paper analysed real samples (43 adulterated and 12 controls) originated from criminal networks dismantled by the Brazilian Police. This fraud consisted of injecting solutions of non-meat ingredients (NaCl, phosphates, carrageenan, maltodextrin) in bovine meat, aiming to increase its water holding capacity. Five physico-chemical variables were determined, protein, ash, chloride, sodium, phosphate. Additionally, infrared spectra were recorded. Supervised classification PLS-DA models were built with each data set individually, but the best model was obtained with data fusion, correctly detecting 91% of the adulterated samples. From this model, a variable selection based on the highest VIPscores was performed and a new data fusion model was built with only one chemical variable, providing slightly lower predictions, but a good cost/performance ratio. Finally, some of the selected infrared bands were specifically associated to the presence of adulterants NaCl, tripolyphosphate and carrageenan.

Calibration transfer from powder mixtures to intact tablets: A new use in pharmaceutical analysis for a known tool.

Calibration transfer is commonly used for spectra obtained in different spectrometers or other conditions. This paper proposed the use of calibration transfer between spectra recorded for the same samples in different physical forms. A new method was developed for the direct determination of nevirapine in solid pharmaceutical formulations based on diffuse reflectance near infrared spectroscopy (NIRS) and partial least squares (PLS). This method was developed with 50 powder mixtures and then, successfully extended to the quantification in intact tablets by using calibration transfer with double window piecewise direct standardization (DWPDS). This chemometric strategy provided good results with a small number of tablet transfer samples, only seven, prepared out of the narrow range of active principle ingredients (API) content around the nominal value of the formulation (100%). The method was fully validated in the working range of 83.0-113.9% of nevirapine and the use of DWPDS allowed to significantly decreasing the root mean square error of prediction (RMSEP) from 4.8% (tablets predicted by a model built with only powder samples) to 2.6%. The range of relative errors decreased from -5.1/8.7% to -4.6/3.3%. Considering that the amount of raw materials demanded for preparing tablets is up to ten times higher than for powder mixtures, this type of application is of particular interest in pharmaceutical analysis. In the context of process analytical technology (PAT), the use of the same multivariate model in different steps of the production is very advantageous, saving time and labor.

Development and analytical validation of a screening method for simultaneous detection of five adulterants in raw milk using mid-infrared spectroscopy and PLS-DA.

This paper proposed a new screening method for the simultaneous detection of five common adulterants in raw cow milk by using attenuated total reflectance (ATR) mid infrared spectroscopy and multivariate supervised classification (partial least squares discrimination analysis - PLSDA). The method was able to detect the presence of the adulterants water, starch, sodium citrate, formaldehyde and sucrose in milk samples containing from one up to five of these analytes, in the range of 0.5-10% w/v. A multivariate qualitative validation was performed, estimating specific figures of merit, such as false positive and false negative rates, selectivity, specificity and efficiency rates, accordance and concordance. The proposed method does not need any sample pretreatment, requires a small amount of sample (30 µL), is fast and simple, being suitable for the control of raw milk in a dairy industry or for the quality inspection of commercialized milk.

Evaluation of transformer insulating oil quality using NIR, fluorescence, and NMR spectroscopic data fusion.

Power transformers are essential components in electrical energy distribution. One of their most important parts is the insulation system, consisting of Kraft paper immersed in insulating oil. Interfacial tension and color are major parameters used for assessing oil quality and the system׳s degradation. This work proposes the use of near infrared (NIR), molecular fluorescence, and (1)H nuclear magnetic resonance (NMR) spectroscopy methods combined with chemometric multivariate calibration methods (Partial Least Squares - PLS) to predict interfacial tension and color in insulating mineral oil samples. Interfacial tension and color were also determined using tensiometry and colorimetry as standard reference methods, respectively. The best PLS model was obtained when NIR, fluorescence, and NMR data were combined (data fusion), demonstrating synergy among them. An optimal PLS model was calculated using the selected group of variables according to their importance on PLS projections (VIP). The root mean square errors of prediction (RMSEP) values of 2.9 mN m(-1) and 0.3 were estimated for interfacial tension and color, respectively. Mean relative standard deviations of 1.5% for interfacial tension and 6% for color were registered, meeting quality control requirements set by electrical energy companies. The methods proposed in this work are rapid and simple, showing great advantages over traditional approaches, which are slow and environmentally unfriendly due to chemical waste generation.

Development and analytical validation of a simple multivariate calibration method using digital scanner images for sunset yellow determination in soft beverages.

This paper proposed a novel methodology for the quantification of an artificial dye, sunset yellow (SY), in soft beverages, using image analysis (RGB histograms) and partial least squares regression. The developed method presented many advantages if compared with alternative methodologies, such as HPLC and UV/VIS spectrophotometry. It was faster, did not require sample pretreatment steps or any kind of solvents and reagents, and used a low cost equipment, a commercial flatbed scanner. This method was able to quantify SY in isotonic drinks and orange sodas, in the range of 7.8-39.7 mg L(-1), with relative prediction errors lower than 10%. A multivariate validation was also performed according to the Brazilian and international guidelines. Linearity, accuracy, sensitivity, bias, prediction uncertainty and a recently proposed tool, the ß-expectation tolerance intervals, were estimated. The application of digital images in food analysis is very promising, opening the possibility for automation.

Chemometric quality inspection control of pyrantel pamoate, febantel and praziquantel in veterinary tablets by mid infrared spectroscopy.

This paper describes the development and validation of a new multivariate calibration method based on diffuse reflectance mid infrared spectroscopy for direct and simultaneous determination of three veterinary pharmaceutical drugs, pyrantel pamoate, praziquantel and febantel, in commercial tablets. The best synergy interval partial least squares (siPLS) model was obtained by selecting three spectral regions, 3715-3150, 2865-2583, and 2298-1733 cm(-1), preprocessed by first derivative and Savitzky-Golay smoothing followed by mean centering. This model was built with five latent variables and provided root mean square errors of prediction (RMSEP) equal or lower than 0.69 mg per 100 mg of powder for the three analytes. The method was validated according the appropriate regulations through the estimate of figures of merit, such as trueness, precision, linearity, analytical sensitivity, bias and residual prediction deviation (RPD). Then, it was applied to three different veterinary pharmaceutical formulations found in the Brazilian market, in a situation of multi-product calibration, since the excipient composition of these commercial products, which was not known a priori, was modeled by an experimental design that scanned the likely content range of the possible constituents. The results were verified with high performance liquid chromatography with diode array detection (HPLC-DAD) and high performance liquid chromatography-tandem mass spectrometry (HPLC-MS/MS) and were in agreement with the predicted values at 95% confidence level. The developed method presented the advantages of being simple, rapid, solvent free, and about ten times faster than the HPLC ones.

Electrospray ionization mass spectrometry and partial least squares discriminant analysis applied to the quality control of olive oil.

Direct infusion electrospray ionization mass spectrometry in the positive ion mode [ESI(+)-MS] is used to obtain fingerprints of aqueous-methanolic extracts of two types of olive oils, extra virgin (EV) and ordinary (OR), as well as of samples of EV olive oil adulterated by the addition of OR olive oil and other edible oils: corn (CO), sunflower (SF), soybean (SO) and canola (CA). The MS data is treated by the partial least squares discriminant analysis (PLS-DA) protocol aiming at discriminating the above-mentioned classes formed by the genuine olive oils, EV (1) and OR (2), as well as the EV adulterated samples, i.e. EV/SO (3), EV/CO (4), EV/SF (5), EV/CA (6) and EV/OR (7). The PLS-DA model employed is built with 190 and 70 samples for the training and test sets, respectively. For all classes (1-7), EV and OR olive oils as well as the adulterated samples (in a proportion varying from 0.5 to 20.0% w/w) are properly classified. The developed methodology required no ions identification and demonstrated to be fast, as each measurement lasted about 3 min including the extraction step and MS analysis, and reliable, because high sensitivities (rate of true positives) and specificities (rate of true negatives) were achieved. Finally, it can be envisaged that this approach has potential to be applied in quality control of EV olive oils.

Development and analytical validation of a multivariate calibration method for determination of amoxicillin in suspension formulations by near infrared spectroscopy.

This paper proposes a new method for determination of amoxicillin in pharmaceutical suspension formulations, based on transflectance near infrared (NIR) measurements and partial least squares (PLS) multivariate calibration. A complete methodology was implemented for developing the proposed method, including an experimental design, data preprocessing by using multiple scatter correction (MSC) and outlier detection based on high values of leverage, and X and Y residuals. The best PLS model was obtained with seven latent variables in the range from 40.0 to 65.0 mg mL(-1) of amoxicillin, providing a root mean square error of prediction (RMSEP) of 1.6 mg mL(-1). The method was validated in accordance with Brazilian and international guidelines, through the estimate of figures of merit, such as linearity, precision, accuracy, robustness, selectivity, analytical sensitivity, limits of detection and quantitation, and bias. The results for determinations in four commercial pharmaceutical formulations were in agreement with the official high performance liquid chromatographic (HPLC) method at the 99% confidence level. A pseudo-univariate calibration curve was also obtained based on the net analyte signal (NAS). The proposed chemometric method presented the advantages of rapidity, simplicity, low cost, and no use of solvents, compared to the principal alternative methods based on HPLC.

Direct determination of trans-resveratrol in human plasma by spectrofluorimetry and second-order standard addition.

Trans-resveratrol (RVT) is an antioxidant found in red grapes and their derivatives, which has been related to the reduction of cardiovascular diseases and cancer incidence. This work developed a new spectrofluorimetric-chemometric method for the direct determination of RVT in human plasma. For each measurement, excitation-emission matrices were obtained from 280 to 360 nm (excitation) and from 380 to 550 nm (emission). The strategy adopted in this work combined data treatment with parallel factor analysis (PARAFAC), for extracting the pure analyte signal, using the standard addition method, which permits determinations in the presence of a strong matrix effect caused by plasma analyte-protein binding. Plasma samples were diluted 10 times and, for each, four standard additions of RVT were performed, in triplicate. A specific PARAFAC model was built for the three replicates of each sample, from three-way arrays formed by five measurements (initial sample plus four additions), 17 excitation wavelengths and 86 emission wavelengths. The best models were selected with four factors and accounted for more than 99.90% of the data variance. The loadings obtained were related to RVT and three interferences. The scores related to the analyte were used for linear regressions and all standard addition curves presented correlation coefficients equal or greater than 0.99. Good results were obtained in the concentration range from 0.10 to 5.00 microg mL(-1), with recoveries between 94.0 and 110.0%. The proposed method was also validated through the estimates of several figures of merit: sensitivity, analytical sensitivity, selectivity, precision, and limits of detection and quantitation.

Direct determination of propranolol in urine by spectrofluorimetry with the aid of second order advantage.

This work presented an application of the second-order advantage provided by parallel factor analysis (PARAFAC) aiming at direct determination of propranolol, a beta-blocker also used as doping agent, in human urine by spectrofluorimetry. The adopted strategy combined the use of PARAFAC, for extraction of the pure analyte signal, with the standard addition method, for a determination in the presence of an individual matrix effect caused by the quenching action of the proteins present in the urine. The urine samples were previously 100 times diluted. For each sample, four standard additions were performed, in triplicates. A specific PARAFAC model was built for each triplicate of each sample, from three-way arrays formed by 231 emission wavelengths, 8 excitation wavelengths and 5 measurements (sample plus 4 additions). The models were built with three factors and always explained more than 99.87% of the total variance. The obtained loadings were related to PRO and two background interferences. The scores related to PRO were used for a linear regression in the standard addition method. The obtained determinations in the PRO concentration range from 5.0 to 20.0 microg ml(-1) provided recoveries between 91.1 and 108.4%.

Combining standard addition method and second-order advantage for direct determination of salicylate in undiluted human plasma by spectrofluorimetry.

Second-order advantage turns possible a determination in the presence of unknown interferences. This work presented an application of the second-order advantage provided by parallel factor analysis (PARAFAC). The aim was the direct determination of salicylic acid (SA), the main product of aspirin degradation, in undiluted human plasma by spectrofluorimetry. The strategy of this analysis combined the use of PARAFAC, for extraction of the pure analyte signal, with the standard addition method, for a determination in the presence of a strong matrix effect caused by the quenching effect of the proteins present in the plasma. For each sample, four standard additions were performed, in triplicates. A specific PARAFAC model was built for each triplicate of each sample, from three-way arrays formed by 436 emission wavelengths, 7 excitation wavelengths and 5 measurements (sample plus 4 additions). In all the cases, the models were built with three factors and explained more than 99.90% of the total variance. The obtained loadings were related to SA and two background interferences. The scores related to SA were used for a linear regression in the standard addition method. Good results were obtained for determinations in the SA concentration range from 3.0 to 24.0mugml(-1), providing errors of prediction between 0.7 and 6.3%.

Direct determination of diclofenac in pharmaceutical formulations containing B vitamins by using UV spectrophotometry and partial least squares regression.

This work proposes a simple and rapid analytical procedure for determination of diclofenac (DCF) in the presence of B vitamins, based on UV measurements and partial least squares (PLS). The interference of thiamine (THI) and pyridoxine (PYR) were modelled using an experimental design constructed in the ranges of 10-50 micromoll(-1) for DCF and THI and 15-75 micromoll(-1) for PYR. The procedure was repeated at five different pH values (between 3 and 6) and the best results were observed at pH 5, presenting a root mean square error of prediction (RMSEP) of 0.80 micromoll(-1) for DCF. The procedure was successfully applied to simultaneous determination of DCF, THI and PYR in synthetic mixtures and in a pharmaceutical formulation that contains a simple excipient (lactose). For determination of a more complex formulation that contains 15 different substances in the excipient, including some UV absorbing ones, the procedure was only able to determine DCF, since the excipient interferences disturbed THI and PYR predictions. Figures of merit, such as selectivity, analytical sensitivity, limit of detection and precision were determined for the DCF prediction model and the determinations were verified by an independent method, HPLC.

N-way PLS applied to simultaneous spectrophotometric determination of acetylsalicylic acid, paracetamol and caffeine.

In this work, a simple and rapid analytical procedure was proposed for simultaneous determination of acetylsalicylic acid (ASA), paracetamol (PRC, also known as acetaminophen) and caffeine (CAF) in pharmaceutical formulations based on multivariate calibration and UV spectrophotometric measurements (210-300 nm). The calibration set was constructed with nine solutions in the concentration ranges from 10.0 to 15.0 microg x ml(-1) for ASA and PRC and from 2.0 to 6.0 microg x ml(-1) for CAF, according to an experimental design. The procedure was repeated at four different pH values: 2.0, 3.0, 4.0 and 5.0. Partial least squares (PLS) models were built at each pH and used to determinate a set of synthetic mixtures. The best model was obtained at pH 5.0. An N-way PLS model was applied to a three-way array constructed using all the pH data sets and enabled better results. This calibration model provided root mean squares errors of prediction (RMSEP) from 11.5 to 35% lower than those obtained with PLS at pH 5.0, depending on the analyte. The results achieved for the determination of these drugs in commercial tablets were in agreement to the values specified by the manufactures and the recovery was between 94.7 and 104.5%.