Ordered Predictors Selection Research Articles

Use of NIR in COVID-19 Screening: Proof of Principles for Future Application.

The COVID-19 pandemic that affected the world between 2019 and 2022 showed the need for new tools to be tested and developed to be applied in global emergencies. Although standard diagnostic tools exist, such as the reverse-transcription polymerase chain reaction (RT-PCR), these tools have shown severe limitations when mass application is required. Consequently, a pressing need remains to develop a rapid and efficient screening test to deliver reliable results. In this context, near-infrared spectroscopy (NIRS) is a fast and noninvasive vibrational technique capable of identifying the chemical composition of biofluids. This study aimed to develop a rapid NIRS testing methodology to identify individuals with COVID-19 through the spectral analysis of swabs collected from the oral cavity. Swab samples from 67 hospitalized individuals were analyzed using NIR equipment. The spectra were preprocessed, outliers were removed, and classification models were constructed using partial least-squares for discriminant analysis (PLS-DA). Two models were developed: one with all the original variables and another with a limited number of variables selected using ordered predictors selection (OPS-DA). The OPS-DA model effectively reduced the number of redundant variables, thereby improving the diagnostic metrics. The model achieved a sensitivity of 92%, a specificity of 100%, an accuracy of 95%, and an AUROC of 94% for positive samples. These preliminary results suggest that NIRS could be a potential tool for future clinical application. A fast methodology for COVID-19 detection would facilitate medical diagnoses and laboratory routines, helping to ensure appropriate treatment.

Characterization and Quantification of Naphthenic Acids in Produced Water by Orbitrap MS and a Multivariate Approach.

Naphthenic acids (NAs) naturally occur in crude oil and its associated produced water, presenting significant challenges, such as corrosion, in refinery apparatus and ecotoxicity in aquatic habitats. This study delineates a multivariate method to quantify NAs in produced water via electrospray ionization coupled with high-resolution Orbitrap mass spectrometry (ESI-Orbitrap MS). By employing liquid-liquid extraction, followed by direct infusion ESI(-)-Orbitrap MS, we characterized and quantified NAs employing a partial least-squares regression (PLS) model enhanced by the ordered predictor selection (OPS) algorithm. Thirty-six produced water samples were utilized, with 24 allocated for calibration and 12 designated for validation. The PLS-OPS model demonstrated notable accuracy in predicting NA concentrations in simulated and actual produced water samples ranging from ∼30 to 300 mg·L-1. This methodology offers a rapid yet robust alternative for quantifying NAs using mass spectrometry augmented by PLS and the OPS. Its significance is underscored by its potential to equip the petroleum industry with a swift and reliable monitoring mechanism for NAs in produced water, thereby aiding in mitigating environmental and operational risks.

Determination of total protein and sugar content in soy-based beverages using variable selection methods applied to ATR-FTIR spectroscopy

This study explores the potential of using ATR-FTIR coupled with partial least squares (PLS) and random forest (RF) for predicting total protein (TP), total sugar (TS), reducing sugars (RS), and non-reducing sugars (NRS) in soy-based beverages (SBBs). Employing variable selection techniques such as interval PLS (iPLS), synergy interval PLS (siPLS), uninformative variable elimination (UVE), ordered predictors selection (OPS), bat algorithm (BA), genetic algorithm (GA), and particle swarm optimization (PSO). The OPS-PLS emerges as the optimal model for TS prediction, yielding an RMSEP of 0.114 wt% and R2p of 0.934. GA-PLS excels in TP, RS, and NRS prediction, achieving RMSEP values of 0.024 wt%, 0.273 wt%, and 0.339 wt%, with corresponding R2p values of 0.880, 0.847, and 0.450. In RF modeling, GA-RF is identified as the superior model for all properties. These findings underscore the potential of regression models and ATR-FTIR as effective and environmentally friendly tools for analyzing SBB composition.

VARIABLE SELECTION METHODS APPLIED IN HTGC DATA TO DETERMINE PHYSICOCHEMICAL PROPERTIES OF CRUDE OILS

High-temperature gas chromatography (HTGC) is an analytical technique employed in the petroleum industry for component separation. By incorporating chemometrics, HTGC data can be effectively utilized to predict various properties of crude oil. However, HTGC chromatograms generate a substantial number of variables, some of which may lack pertinent chemical information. Consequently, employing variable selection methods becomes crucial to reduce the number of variables and enhance the predictive capability of calibration models. In this study, the interval partial least squares (iPLS), synergy interval partial least squares (siPLS), and ordered predictors selection (OPS) methods were applied for variable selection to construct linear regression models. The main objective was to investigate the potential of these methods in predicting eight properties of crude oil: American Petroleum Institute (API) gravity, standardized kinematic viscosity at 50 °C (VISp), flash point (FP), Reid vapor pressure (RVP), micro carbon residue (MCR), saturates (SAT), aromatics (ARO), and polar (POL) content. While all variable selection methods yielded satisfactory results, the OPS-PLS regression models consistently exhibited the best performance in estimating these properties, achieving root mean squared error of prediction (RMSEP) values of 1.244 for API, 0.029 for VISp, 15.356 °C for FP, 0.324 kPa for RVP, 0.629 wt% for MCR, 3.691 wt% for SAT, 2.939 wt% for ARO, and 3.374 wt% for POL. Variable selection demonstrated remarkable effectiveness, significantly improving the accuracy of the models, and allowing for the creation of concise models with a focused set of variables.

Predicting ruminal degradability and chemical composition of corn silage using near-infrared spectroscopy and multivariate regression.

The aim of this study was to develop and validate regression models to predict the chemical composition and ruminal degradation parameters of corn silage by near-infrared spectroscopy (NIR). Ninety-four samples were used to develop and validate the models to predict corn silage composition. A subset of 23 samples was used to develop and validate models to predict ruminal degradation parameters of corn silage. Wet chemistry methods were used to determine the composition values and ruminal degradation parameters of the corn silage samples. The dried and ground samples had their NIR spectra scanned using a poliSPECNIR 900-1700 model NIR sprectrophotometer (ITPhotonics S.r.l, Breganze, IT.). The models were developed using regression by partial least squares (PLS), and the ordered predictor selection (OPS) method was used. In general, the regression models obtained to predict the corn silage composition (P>0.05), except the model for organic matter (OM), adequately estimated the studied properties. It was not possible to develop prediction models for the potentially degradable fraction in the rumen of OM and crude protein and the degradation rate of OM. The regression models that could be obtained to predict the ruminal degradation parameters showed correlation coefficient of calibration between 0.530 and 0.985. The regression models developed to predict CS composition accurately estimated the CS composition, except the model for OM. The NIR has potential to be used by nutritionists as a rapid prediction tool for ruminal degradation parameters in the field.

A Novel Variable Selection Method Based on Ordered Predictors Selection and Successive Projections Algorithm for Predicting Gastrodin Content in Fresh Gastrodia elata Using Fourier Transform Near-Infrared Spectroscopy and Chemometrics.

Gastrodin is one of the most important biologically active components of Gastrodia elata, which has many health benefits as a dietary and health food supplement. However, gastrodin measurement traditionally relies on laboratory and sophisticated instruments. This research was aimed at developing a rapid and non-destructive method based on Fourier transform near infrared (FT-NIR) to predict gastrodin content in fresh Gastrodia elata. Auto-ordered predictors selection (autoOPS) and successive projections algorithm (SPA) were applied to select the most informative variables related to gastrodin content. Based on that, partial least squares regression (PLSR) and multiple linear regression (MLR) models were compared. The autoOPS-SPA-MLR model showed the best prediction performances, with the determination coefficient of prediction (Rp2), ratio performance deviation (RPD) and range error ratio (RER) values of 0.9712, 5.83 and 27.65, respectively. Consequently, these results indicated that FT-NIRS technique combined with chemometrics could be an efficient tool to rapidly quantify gastrodin in Gastrodia elata and thus facilitate quality control of Gastrodia elata.

UV-Vis spectroscopy and one-class modeling for the authentication of the geographical origin of green coffee beans from Cerrado Mineiro, Brazil

Green coffee beans produced in the Cerrado Mineiro region have a Protected Designation of Origin (PDO) certificate. This guarantees that its coffees are among Brazil's most appreciated and valued. This study aimed to develop an authentication method to characterize and differentiate coffee beans from this region. Ultraviolet-visible spectra of 130 green coffee bean samples were used for one-class modeling with SIMCA (soft independent modeling of class analogy), DD-SIMCA (data-driven SIMCA), and OCPLS (one-class partial least squares). Ordered predictors selection (OPS) variable selection was applied to simplify multivariate models and improve their performance. The interpretation of developed models was performed by evaluating the parameter modeling power for variables of SIMCA models. It was possible to assign the most predictive variables to trigonelline and chlorogenic acids as responsible for discriminating coffee beans from Cerrado Mineiro from coffee beans of their three main competitor regions: Caparaó, Mogiana, and Sul de Minas. Finally, methodologies as developed in this work, based on one-class modeling methods, can be proposed to authenticate other coffee-producing regions, food matrices, and agricultural products.

Study of coffee sensory attributes by ordered predictors selection applied to 1H NMR spectroscopy

The determination of sensory attributes of coffee samples through trained cupping tasters is one of the main methodologies for coffee quality control. However, this analysis may suffer some distortions due to different ways of interpreting quality from the tasters' perspective. Thus, other, and complementary have been developed as alternative ways to determine sensory attributes. Hydrogen nuclear magnetic resonance (1H NMR) is a technique that allows the chemical study of compounds present in organic matrices. Through 1H NMR it is possible to predict a sensory attribute using chemometric methods. However, 1H NMR generates many variables per sample, making it necessary to use variable selection methods to find the most useful information. Ordered predictors selection (OPS) is a variable selection method developed to search, sort, and select the variables in order of importance. In this study, six sensory attributes from 42 coffee samples were predicted by partial least squares (PLS) and support vector regression (SVR) methods and OPS as variable selection in 1H NMR data set. The integration of areas of the 1H NMR spectra and the full data sets were also used to compare the prediction models. OPS-PLS model was the best strategy to predict aftertaste, balance, and fragrance with root mean squared error of prediction (RMSEP) of 0.24, 0.35, and 0.20, respectively. The coefficients of determination for prediction (R2p) of these models were 0.71, 0.60, and 0.77, respectively. OPS-SVR predominated as the best strategy to predict acidity, body, and flavor with RMSEP of 0.23, 0.21, and 0.22, and R2p of 0.89, 0.75, and 0.83, respectively. None of the models chosen as the best showed the presence of trends in residuals above 5% of significance. Variable selection by OPS can improve the prediction of regression models.

Synthesis, antiproliferative and 4D-QSAR studies of thiadiazole derivatives bearing acrylamide moiety as EGFR inhibitors

ABSTRACT As a target for clinical anti-cancer treatment, epidermal growth factor receptor (EGFR) exhibits its over-expression on various tumour cells and is associated with the development of a variety of human cancers. Herein, we described the synthesis, antiproliferative activity assay and 4D-QSAR studies of thiadiazole derivatives bearing acrylamide moiety as EGFR inhibitors. Compared with Gefitinib, some of the target compounds have excellent antiproliferative activities against EGFR-expressed A431 cell line. The robust and reliable 4D-QSAR was constructed using comparative distribution detection algorithm, ordered predictors selection and genetic algorithm method, and the following acceptable statistics are shown: r 2 = 0.82, Q 2 LOO = 0.67, Q 2 LMO = 0.61, r 2 Pred = 0.78.

Screening method for the characterization of anabolic steroids seized in Brazil using paper spray mass spectrometry and chemometric tools.

This paper reports the use of paper spray mass spectrometry (PS-MS) combined with chemometric models to analyze seized samples of anabolic steroids. Because many forensic laboratories typically demand high-throughput analysis for this type of sample, we developed a quicker and simpler alternative analytical method for routine analysis with minimal sample preparation. Oily samples (n = 39) resulting from seizures carried out by Brazilian Federal and State Police units were selected for this study. These samples were analyzed by PS-MS in the positive ion mode and full scan (50-1000 m/z), providing spectra containing patterns of the respective active ingredients present in each product. A principal component analysis (PCA) model was built, which discriminated samples mainly according to their active ingredients and allowed to detect and characterize some cases of product counterfeiting. The variable selection method ordered predictors selection was employed jointly with PCA to improve sample cluster separation and to provide model simplification. The final PCA model was built with three principal components and using only 28 spectral variables. This model accounted for 69.82% of the variance and discriminated samples according to their specific active ingredients.

Non-Invasive Method to Predict the Composition of Requeijão Cremoso Directly in Commercial Packages Using Time Domain NMR Relaxometry and Chemometrics

Low Field Time-Domain Nuclear Magnetic Resonance (TD-NMR) relaxometry was used to determine moisture, fat, and defatted dry matter contents in “requeijão cremoso” (RC) processed cheese directly in commercial packaged (plastic cups or tubes with approximately 200 g). Forty-five samples of commercial RC types (traditional, light, lactose-free, vegan, and fiber) were analyzed using longitudinal (T1) and transverse (T2) relaxation measurements in a wide bore Halbach magnet (0.23 T) with a 100 mm probe. The T1 and T2 analyses were performed using CWFP-T1 (Continuous Wave Free Precession) and CPMG (Carr-Purcell-Meiboom-Gill) single shot pulses. The scores of the principal component analysis (PCA) of CWFP-T1 and CPMG signals did not show clustering related to the RC types. Optimization by variable selection was carried out with ordered predictors selection (OPS), providing simpler and predictive partial least squares (PLS) calibration models. The best results were obtained with CWFP-T1 data, with root-mean-square errors of prediction (RMSEP) of 1.38, 4.71, 3.28, and 3.00% for defatted dry mass, fat in the dry and wet matter, and moisture, respectively. Therefore, CWFP-T1 data modeled with chemometrics can be a fast method to monitor the quality of RC directly in commercial packages.

Classification of sugarcane genotypes susceptible and resistant to the initial attack of sugarcane borer Diatraea saccharalis using epicuticular wax composition

Identifying compounds present in the sugarcane epicuticular wax and using these compounds to classify the genotypes susceptible and resistant to the initial attack of sugarcane borer (Diatraea saccharalis) was the aim of this study. A greenhouse experiment was performed in a factorial scheme with and without borer infestation using genotypes previously characterized as resistant or susceptible in field-based experiments. Sugarcane whorls of six-month-old plants were collected before (BI) and after (AI) 72 h of sugarcane borer infestation. The sugarcane epicuticular wax was extracted in both times, i.e., BI and AI and its chemical composition was assessed by gas chromatography coupled to mass spectrometry (GC-MS). Twenty-five compounds were identified for both BI and AI. Classification models were built using partial least squares for discriminant analysis (PLS-DA) and linear discriminant analysis (LDA). Variable selection methods were used to improve the classification models. Ordered predictors selection for discriminant analysis (OPSDA) selected compounds that correctly classified all the test samples before borer infestation (Error = 0.000), and exhibited the most suitable classification parameters for the test set after borer infestation (Error = 0.111). The C30 pentacyclic triterpene friedelin and a high alcohol/aldehyde ratio were associated with the classification of resistant genotypes. Our findings have applicability in developing a screening methodology for breeding programs interested in identifying genotypes resistant to the initial feeding of sugarcane borer.

Dehydration as a Tool to improve predictability of sugarcane juice carbohydrates using near-infrared spectroscopy based PLS models

The aim of this work was to study dehydration as a way to improve the prediction of sucrose, glucose, and fructose in sugarcane juice using near-infrared (NIR) spectroscopy and partial least squares (PLS) regression models. The temperature, time, and sample volume involved in the dehydration process were optimized using design of experiments. Six different sample supports were assessed, being the thick couche paper the best support. NIR spectra from liquid (LSJ) and dehydrated sugarcane juice (DSJ) were obtained. Sucrose, glucose, and fructose in LSJ were analyzed using high-performance liquid chromatography with an evaporative light scattering detector (HPLC-ELSD). Sucrose, glucose, and fructose ranged from 99.29 to 249.27 ​mg/mL, 5.96–14.94 ​mg/mL and 3.99–16.10 ​mg/mL. PLS models were built using the sugars content and NIR spectra collected from a benchtop and a portable instrument. Ordered predictors selection (OPS) was applied to select the most informative variable. The results indicated better predictions for all sugars using the DSJ for both instruments, being the benchtop statistically better than the portable instrument. On the benchtop instrument, the PLS-OPS models presented root mean square error of prediction (RMSEP) respectively for sucrose, glucose, and fructose 7.98, 0.82, and 1.00 ​mg/mL using the DSJ against 12.75, 1.00, and 1.35 ​mg/mL using the LSJ. For the portable instrument, the RMSEP were respectively 15.90, 1.18, and 1.65 ​mg/mL using DSJ against 23.23, 1.40, and 2.08 ​mg/mL using LSJ. To sum up, the dehydration approach showed to be a great technique to improve the predictability of PLS-OPS models for sugarcane juice sugars using NIR spectra by removing the water and concentrating the analytes.

Determination of gross calorific value in crude oil by variable selection methods applied to 13C NMR spectroscopy

Gross Calorific Value (GCV) is one of the properties to assess the quality and value of fuel in the oil industry, but the standard method is laborious. Regression models built with nuclear magnetic resonance (13C NMR) data make it possible to estimate different physicochemical properties of oils. However, its adversity is the enormous amount of chemical information produced in a single spectrum. And not all variables contribute in the model. With variable selection methods (VS), we can find the information that has the highest correlation with the property of interest. In our study, we used different methods applied to 13C NMR data of 145 Brazilian crude oil samples with GCV ranging from 41.5 to 47 MJ∙kg−1. For variable selection we used genetic algorithm (GA), variable importance in projection (VIP), uninformative variable elimination (UVE), angular search algorithm with variance inflation factor (ASA-VIF), competitive adaptive reweighted sampling (CARS), synergy interval partial least squares (siPLS), interval partial least squares (iPLS), subwindow permutation analysis (SPA), ordered predictors selection (OPS) and particle swarm optimization (PSO). Also, we used orthogonal projection to latent structures (OPLS) and partial least squares (PLS) with full spectra (PLS-full). All models using variable selection obtained lower root mean squared error of prediction (RMSEP) compared to the model based on full range (PLS-full). PSO was the most accurate model with RMSEP of 0.152 MJ∙kg−1. PSO selected the paraffinic region (0 to 4.43 ppm), totalizing 701 variables. Statistical analysis showed no trends in residues above 5% of significance for the best models.

Predicting oil content in ripe Macaw fruits (Acrocomia aculeata) from unripe ones by near infrared spectroscopy and PLS regression

A method for early quantification of unripe macaw fruits oil content using near-infrared spectroscopy (NIR) and partial least squares (PLS) is presented. After harvest, the fruit takes about 30 days to reach its maximum oil accumulation. The oil content was quantified thirty days after harvest using Soxhlet extraction. PLS models were built using NIR spectra of shell obtained five days after harvest (Shell5). The Shell5 model was compared with models built using NIR spectra of the shell (Shell30) and mesocarp thirty days after harvest (Pulp30). Ordered predictors selection was used to select the most informative variables. The best models presented root mean square error of prediction and correlation coefficient of prediction of 4.87% and 0.89 for Shell5; 5.83% and 0.85 for Shell30; 4.76% and 0.92 for Pulp30. Thus, the anticipated prediction of oil content could reduce the time and costs of macaw palm quality control and storage.

Particle swarm optimization and ordered predictors selection applied in NMR to predict crude oil properties

We used spectral information from nuclear magnetic resonance (1H NMR and 13C NMR) of nearly 150 Brazilian crude oil samples to predict some physicochemical properties. We build models to estimate API gravity (API), standardized kinematic viscosity at 50 °C (VISst), heat combustion value (HCV), total acid number (TAN), saturates (SAT), aromatics (ARO), resins (RES) and asphaltenes (ASF) content. To obtain accurate models, particle swarm optimization (PSO) and ordered predictors selection (OPS) were applied as variable selection techniques coupled to partial least squares (PLS) regression. PSO-PLS and OPS-PLS hybrid models presented higher predictive capacity than PLS regression models. We were able to find the most relevant signal areas of the NMR spectra for each property. The best results of SAT, ARO and RES content were obtained with PSO-PLS of 13C NMR dataset with root mean squared error of prediction (RMSEP) of 4.54, 2.85 and 4.08 (wt%), respectively. For API, VISst and TAN the RMSEP were equal to 0.74 (API), 0.02 and 0.16 (mg KOH·g−1), respectively, using OPS-PLS method and 1H NMR dataset. The more accurated models for ASF contents and HCV were built with PSO-PLS of 1H NMR, with RMSEP of 0.59 (wt%) and PSO-PLS of 13C NMR with RMSEP of 0.64 (MJ·kg−1), respectively. However, these two properties presented a coefficient of determination for the prediction set (R2p) lower than 0.75, which means they are not very well adjusted to the regression model. Furthermore, using 13C NMR dataset, OPS-PLS models has shown similar results than PSO-PLS models, for some properties.

In silico Risk Assessment Studies of New Psychoactive Substances Derived from Amphetamines and Cathinones

The amount and variety of new psychoactive substances (NPS) are expanding, and there are difficulties in assessing their risks. In this regard, in silico methods are potentially useful to predict NPS properties faster and at a lower cost. In this work a quantitative structure-activity relationship (QSAR) model was used to verify the risk of drugs derived from amphetamines and cathinones. A dataset of 26 derivatives with in vitro affinity for norepinephrine transporter (NET) was selected. To ensure reproducibility of the results, only geometric molecular descriptors (AM1 (Austin model 1) level) obtained from the platform ChemDes and ordered predictors selection (OPS) were used. The model presents good internal statistics (n = 23; coefficient of determination (R2) = 0.914). The small number of samples was divided into seven training sets (n = 17) and seven test sets (n = 6). The average R2pred = 0.754 showed that the model has good predictive capacity. Based on the tests, this model can accurately predict the risk range of three previously selected derivatives: methedrone (low), ethcathinone (medium), and methamphetamine (high), even when only data referring to NET are employed. We used these data to create a simple free program in Java that focuses on the risk assessment of recreational drugs belonging to this class of compounds.

Comprehensive new approaches for variable selection using ordered predictors selection

New strategies of ordered predictors selection (OPS) were developed in this work, making this method more versatile and expanding its worldwide use and applicability. OPS is a recognized method to select variables in multivariate regression and is used by analytical chemists and chemometrists. It shows high ability to improve the prediction of models after the selection of a few and important variables. At the core of OPS is sorting variables from informative vectors and systematically investigating the regression models to identify the most relevant set of variables by comparing the cross-validation parameters of the models. Nevertheless, the first version of the OPS method performs variable selection using only one informative vector at a time and is limited to just one variable selection run. Then, three new strategies were proposed. First, an automatic method was developed to perform variable selection using several informative vectors and their combinations. Second, the feedback OPS is presented, in this new strategy the pre-selected variables would return to a new selection. Last, a method to apply OPS in full array subdivisions called OPS intervals was established. Initially, the new strategies were applied in the six datasets used in the original OPS paper to compare the prediction performance with the new OPS algorithms. After that, twelve new datasets were used to test and compare the new OPS approaches with other variable selection methods, genetic algorithm (GA), the interval successive projections algorithm for PLS (iSPA), and recursive weighted partial least squares (rPLS). The new OPS approaches outperformed the first OPS version and the other variable selection methods. Results showed that in addition to greater predictive capacity, the accuracy in the selection of expected variables is highly superior with the new OPS approaches. Overall, the new OPS provided the best set of selected variables to build more predictive and interpretative regression models, proving to be efficient for variable selection in different types of datasets.

Optimizing wavelength selection by using informative vectors for parsimonious infrared spectra modelling

Infrared spectroscopy has been widely adopted by various agricultural research. The typical spectra variables contain thousands of wavelengths. These large number of spectra variables often contribute to collinearity, and redundancies rather than relevant information. Variable selection of the predictors is an important step to create a robust calibration model from these spectra data. This paper presents an algorithm for spectra variable selection based on a combination of informative vectors and an ordered predictor selection (OPS) approach with an exponentially decreasing function (EDF) selection. Informative vectors are features derived from statistical principles that can be used to describe the relationship between the dependent variables and the predictors (spectra). The informative vectors analysed include regression coefficient vector (b), variable influence on projection (V), residual vector (S), net analyte signal vector (Na), linear correlation vector (COR), biweight mid-correlation vector (BIC), mutual information based on adjacency matrix (AMI), covariance procedures matrix (COV). These eight informative vectors can be joined in pairs and become 22 combination vectors. This approach was tested with near-infrared soil spectra for predicting the properties of pH, clay and sand content, cation exchange capacity (CEC), and total carbon content. This example used the Cubist regression tree and partial least squares regression (PLSR) models for calibration. By utilizing the subset of the spectra (retaining those that are significant based on the absolute values of the informative vectors), the regression models were still able to enhance the prediction capability. Overall, the PLSR model performed better than the Cubist model. The informative vector b (and its combinations) and S (and its combinations) were found to be the ones that provide the most accurate predictions for this dataset. Although the performance of the subset model does not perform better than the full spectra model, the number of wavelengths variable used in the model is significantly reduced to, on average, 25%.

Combining mid infrared spectroscopy and paper spray mass spectrometry in a data fusion model to predict the composition of coffee blends

This paper describes a robust multivariate model for quantifying and characterizing blends of Robusta and Arabica coffees. At different degrees of roasting, 120 ground coffee blends (0.0–33.0%) were formulated. Spectra were obtained by two different techniques, attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy and paper spray mass spectrometry (PS-MS). Partial least squares (PLS) models were built individually with the two types of spectra. Nevertheless, better predictions were obtained by low and medium-level data fusion, taking advantage from the synergy between these two data sets. Data fusion models were improved by variable selection, using genetic algorithms (GA) and ordered predictors selection (OPS). The smallest prediction errors were provided by OPS low-level data fusion model. The number of variables used for regression was reduced from 2145 (full spectra) to 230. Model interpretation was performed by assigning some of the selected variables to specific coffee components, such as trigonelline and chlorogenic acids.