The amounts of data collected from each sample of e.g. chemical or biological materials have increased by orders of magnitude since the beginning of the 20th century. Furthermore, the number of ways to collect data from observations is also increasing. Such configurations with several massive data sets increase the demands on the methods used to analyse them. Methods that handle such data are called multiblock methods and they are the topic of this thesis. Data collected from advanced analytical instruments often contain variation from diverse mutually independent sources, which may confound observed patterns and hinder interpretation of latent variable models. For this reason, new methods have been developed that decompose the data matrices, placing variation from different sources of variation into separate parts. Such procedures are no longer merely pre-processing filters, as they initially were, but have become integral elements of model building and interpretation. One strain of such methods, called OPLS, has been particularly successful since it is easy to use, understand and interpret. This thesis describes the development of a new multiblock data analysis method called OnPLS, which extends the OPLS framework to the analysis of multiblock and path models with very general relationships between blocks in both rows and columns. OnPLS utilises OPLS to decompose sets of matrices, dividing each matrix into a globally joint part (a part shared with all the matrices it is connected to), several locally joint parts (parts shared with some, but not all, of the connected matrices) and a unique part that no other matrix shares. The OnPLS method was applied to several synthetic data sets and data sets of “real” measurements. For the synthetic data sets, where the results could be compared to known, true parameters, the method generated global multiblock (and path) models that were more similar to the true underlying structures compared to models without such decompositions. I.e. the globally joint, locally joint and unique models more closely resembled the corresponding true data. When applied to the real data sets, the OnPLS models revealed chemically or biologically relevant information in all kinds of variation, effectively increasing the interpretability since different kinds of variation are distinguished and separately analysed. OnPLS thus improves the quality of the models and facilitates better understanding of the data since it separates and separately analyses different kinds of variation. Each kind of variation is purer and less tainted by other kinds. OnPLS is therefore highly recommended to anyone engaged in multiblock or path model data analysis.

This thesis presents a novel bioinformatic methodology denoted the bio-chemometric approach. The methodology is designed for generation of detailed descriptions and predictions of biomolecular interactions. It is based on multivariate analysis of the sensitivity of a biomolecular interaction to multiple minor changes in the experimental conditions. In this work, either the chemical environment where the interaction takes place, or the molecular structure of one of the interacting molecules, was varied. The sensitivity of the interaction to the performed variations was presented as a vector called the sensitivity fingerprint. The bio-chemometric approach was tested on several biomolecular interactions. Useful descriptions of the interactions were obtained by measuring binding kinetics for each interaction in 12-20 different buffers and correlating buffer composition to binding kinetics. The obtained chemical sensitivity fingerprints were reproducible, significantly different and showed a weak correlation to binding site properties for the tested interactions. The results indicate that the fingerprints contained useful information about the binding site. The predictive ability of the bio-chemometric approach was tested on two different biomolecular interactions where one of the binding partners was slightly modified into multiple analogues by amino acid exchanges. In one example, interactions of 18 peptide analogues with an antibody gave data that could be used for accurate prediction of the dissociation rates of novel analogues. Reliable predictions of binding kinetics and affinity were also obtained for single domain camel antibody analogues binding to a protein antigen. By using the three-dimensional structure of camel antibodies and data obtained using the bio-chemometric approach, even the importance of non-exchanged amino acids for the binding could be estimated. The bio-chemometric approach can potentially improve the development of peptides and proteins for therapeutic and diagnostic use. It is suggested to be valid for general use in biochemistry.

Membrane filtration of protein solutions is influenced by a wide range of processing and physicochemical conditions. Monitoring and optimizing membrane filtration may have advantages for achieving, in a cost effective manner, improved bioproduct purification and membrane performance which is relevant to pharmaceutical and biochemical applications. The motivation of this work was to examine the feasibility of applying two-dimensional fluorescence spectroscopy in conjunction with chemometric techniques for monitoring and possibly optimizing the performance of membrane processes. Preliminary work focused on assessing the use of multivariate calibration tools in conjunction with the sensitivity of intrinsic protein fluorescence towards changes in environmental conditions was to predict protein concentration and aggregation behavior. A model protein, β-lactoglobulin (β-LG), was used as a first simple case scenario. Results showed very good agreement between the fluorescence based predictions and measurements obtained by HPLC and gravimetric analysis regardless of the conditions. PLS analysis of excitation-emission matrices revealed unique spectral fingerprints that are most likely associated with the heat-induced denaturation and aggregation. Standard Normal Variate, a signal preprocessing and filtering tool, was shown to have a significant effect on enhancing the predictive accuracy and robustness of the PLS model as it reduced the effect of instrumental noise. The methodology was then extended to a two-component protein system consisting of α- lactlalbumin (α-LA) and β-lactoglobulin (β-LG). The process of thermal induced aggregation of β-LG and α-LA protein in mixtures, which involves the disappearance of native-like proteins, was studied under various treatment conditions including different temperatures, pH, total initial protein concentration and proportions of α-LA and β-LG. A Partial Least Squares (PLS) regression algorithm was used to correlate the concentrations of α-LA and β-LG to the fluorescence spectra obtained for mixtures.The results illustrated that multivariate models could effectively deconvolute multiwavelength fluorescence spectra collected for the protein mixtures and thereby provide a fairly accurate quantification of respective native-like α-LA and β-LG despite the significant overlap between their emission profiles. It was also demonstrated that a PLS model could be used as a black-box prediction tool for estimating protein aggregation when combined with simple mass balances. Ultrafiltration experiments of the whey protein isolate solutions were carried out in dead-end filtration mode and fluorescence measurements of permeate and retentate solutions were acquired in synchronous scanning mode using a fiber optic probe. By implementing a dilution strategy for the retentate side, the fluorescence based PLS model encompassed a low protein concentration range where fluorescence was not expected to be significantly influenced by concentration-dependent interferences. It was also demonstrated that synchronous spectra can provide good predictions and consequently the use of the full spectrum may not be necessary for monitoring with corresponding savings in acquisition time. Membrane performance variables that are difficult to measure, such as individual protein transmission and membrane selectivity could be estimated directly from fluorescence-based predictions of protein concentrations in the retentate and permeate streams. Multiwavelength light scattering spectra, acquired using the fiber optic probe, were shown to be a useful indicator for protein self-association behavior, which is known to influence the membrane filtration. High fouling potential were observed for protein solutions that exhibited significant Rayleigh scattering. A predictive PLS model for estimating protein aggregation from Rayleigh scattering measurements was developed and it was tested by using molecular weight experimental values obtained from the literature. Although this comparison was only partial due to the limited amount of molecular weight data available, the findings verified the possibility of estimating the aggregate size from multiwavelength Rayleigh scattering spectra acquired using a conventional spectrofluorometer. Thus, the results implied that both intrinsic fluorescence and light scattering multiwavelength measurements could provide complementary information about the filtration process.

Membrane filtration of protein solutions is influenced by a wide range of processing and physicochemical conditions. Monitoring and optimizing membrane filtration may have advantages for achieving, in a cost effective manner, improved bioproduct purification and membrane performance which is relevant to pharmaceutical and biochemical applications. The motivation of this work was to examine the feasibility of applying two-dimensional fluorescence spectroscopy in conjunction with chemometric techniques for monitoring and possibly optimizing the performance of membrane processes. Preliminary work focused on assessing the use of multivariate calibration tools in conjunction with the sensitivity of intrinsic protein fluorescence towards changes in environmental conditions was to predict protein concentration and aggregation behavior. A model protein, β-lactoglobulin (β-LG), was used as a first simple case scenario. Results showed very good agreement between the fluorescence based predictions and measurements obtained by HPLC and gravimetric analysis regardless of the conditions. PLS analysis of excitation-emission matrices revealed unique spectral fingerprints that are most likely associated with the heat-induced denaturation and aggregation. Standard Normal Variate, a signal preprocessing and filtering tool, was shown to have a significant effect on enhancing the predictive accuracy and robustness of the PLS model as it reduced the effect of instrumental noise. The methodology was then extended to a two-component protein system consisting of α- lactlalbumin (α-LA) and β-lactoglobulin (β-LG). The process of thermal induced aggregation of β-LG and α-LA protein in mixtures, which involves the disappearance of native-like proteins, was studied under various treatment conditions including different temperatures, pH, total initial protein concentration and proportions of α-LA and β-LG. A Partial Least Squares (PLS) regression algorithm was used to correlate the concentrations of α-LA and β-LG to the fluorescence spectra obtained for mixtures.The results illustrated that multivariate models could effectively deconvolute multiwavelength fluorescence spectra collected for the protein mixtures and thereby provide a fairly accurate quantification of respective native-like α-LA and β-LG despite the significant overlap between their emission profiles. It was also demonstrated that a PLS model could be used as a black-box prediction tool for estimating protein aggregation when combined with simple mass balances. Ultrafiltration experiments of the whey protein isolate solutions were carried out in dead-end filtration mode and fluorescence measurements of permeate and retentate solutions were acquired in synchronous scanning mode using a fiber optic probe. By implementing a dilution strategy for the retentate side, the fluorescence based PLS model encompassed a low protein concentration range where fluorescence was not expected to be significantly influenced by concentration-dependent interferences. It was also demonstrated that synchronous spectra can provide good predictions and consequently the use of the full spectrum may not be necessary for monitoring with corresponding savings in acquisition time. Membrane performance variables that are difficult to measure, such as individual protein transmission and membrane selectivity could be estimated directly from fluorescence-based predictions of protein concentrations in the retentate and permeate streams. Multiwavelength light scattering spectra, acquired using the fiber optic probe, were shown to be a useful indicator for protein self-association behavior, which is known to influence the membrane filtration. High fouling potential were observed for protein solutions that exhibited significant Rayleigh scattering. A predictive PLS model for estimating protein aggregation from Rayleigh scattering measurements was developed and it was tested by using molecular weight experimental values obtained from the literature. Although this comparison was only partial due to the limited amount of molecular weight data available, the findings verified the possibility of estimating the aggregate size from multiwavelength Rayleigh scattering spectra acquired using a conventional spectrofluorometer. Thus, the results implied that both intrinsic fluorescence and light scattering multiwavelength measurements could provide complementary information about the filtration process.

In this research, physicochemical changes that took place during roast processing of coffee beans using fluidized air roaster were studied. The results showed that high-temperature-short-time resulted in higher moisture content, higher pH value, higher titratable acidity, higher porous structure in the bean cell tissues, and also produced more aldehydes, ketones, aliphatic acids, aromatic acids, and caffeine than those processed at low-temperature-long-time process. Fourier transform infrared (FTIR) spectroscopy and chemometric analysis showed that clusters for principal components score plots of ground coffee, extracted by a mixture of equal volume of ethyl acetate and water, were well separated. The research indicated that variations in IR-active components in the coffee extracts due to different stages of roast, roasting profiles, and geographical origins can be evaluated by the FTIR technique. / Natural Sciences and Engineering Research Council of Canada (NESRC) and Mother Parkers Tea & Coffee Inc.

Functional data analysis (FDA) is a demonstrably effective, practical, and powerful method of data analysis, yet it remains virtually unheard of outside of academic circles and has almost no exposure to industry. FDA adds to the milieu of statistical methods by treating functions of one or more independent variables as data objects, analogous to the way in which discrete points are the data objects we are familiar with in conventional statistics. The first step in functional analysis is to “functionalize” the data, or convert discrete points into a system represented most times by continuous functions. Choosing the type of functions to use is data-dependent and often straightforward – for example, Fourier series lend themselves well to periodic systems, while splines offer great flexibility in approximating more irregular trends, such as chemical spectra. This work explores the question of how B-splines can be rapidly and reliably used to denoised infrared chemical spectra, a difficult problem not only because of the many parameters involved in generating a spline fit, but also due to the disparate nature of spectra in terms of shape and noise intensity. Automated selection of spline parameters is required to support high-throughput analysis, and the heteroscedastic nature of such spectra presents challenges for existing techniques. The heuristic knot placement algorithm of Li et al. (2005) for 1D object contours is extended to spectral fitting by optimizing the denoising step for a range of spectral types and signal/noise ratios, using the following criteria: robustness to types of spectra and noise conditions, parsimony of knots, low computational demand, and ease of implementation in high-throughput settings. Pareto-optimal filter configurations are determined using simulated data from factorial experimental designs. The improved heuristic algorithm uses wavelet transforms and provides improved performance in robustness, parsimony of knots and the quality of functional regression models used to correlate real spectral data with chemical composition. In practical applications, functional principal component regression models yielded similar or significantly improved results when compared with their discrete partial least squares counterparts. / Thesis (Master, Chemical Engineering) -- Queen's University, 2012-10-01 20:18:31.119

Cleaning validation, in which cleaned surfaces are analysed for residual material, is an important process in pharmaceutical manufacturing and research facilities. Current procedures usually consist of either swab or rinse-water sampling followed by analysis of the samples. The analysis step is typically either rapid but unselective (conductivity, pH, total organic carbon, etc.), or selective but time-consuming (HPLC). This thesis describes the development of an in situ surface-spectroscopic analysis that removes the need for swab sampling and is both rapid and selective. This method has the potential to complement existing analyses to increase the efficiency of cleaning-validation protocols. The spectrometric system consists of a Fourier-transform infrared (FTIR) spectrometer coupled to a fibre-optic grazing-angle reflectance probe, and allows the measurement of infrared reflection-absorbance spectra (IRRAS) from flat surfaces in ~10 s. Multivariate chemometric methods, such as partial least squares (PLS) regression, are used to exploit the high information content of infrared spectra to obtain selective analyses without physical separation of the analyte or analytes from whatever interfering species may be present. Multivariate chemometric models require considerably more effort for calibration and validation than do traditional univariate techniques. This thesis details suitable methods for preparing calibration standards by aerosol deposition, optimising and validating the model by cross- and test-set validation, and estimating the uncertainty by resampling and formula-based approaches. Successful calibration models were demonstrated for residues of acetaminophen, a model active pharmaceutical ingredient (API), on glass surfaces. The root-mean-square error of prediction (RMSEP) was ~0.07 µg cm⁻². Simultaneous calibration for acetaminophen and aspirin, another API, gave a similar RMSEP of 0.06 µg cm⁻² for both compounds, demonstrating the selectivity of the method. These values correspond to detection limits of ~0.2 µg cm⁻², well below the accepted visual detection limit of ~1-4 µg cm⁻². The sensitivity of the method with a stainless steel substrate was found to depend strongly on the surface finish, with highly polished surfaces giving more intense IRRAS. RMSEP values of 0.04- 0.05 µg cm⁻² were obtained for acetaminophen on stainless steel with three different finishes. For this system, severe nonlinearity was encountered for loadings 1.0 µg cm⁻². From the results presented in this thesis, it is clear that IRRAS has potential utility in cleaning validation as a complement to traditional techniques.

Thesis (Ph. D.)--University of Canterbury, 2006. / Typescript (photocopy). "September 2006." Includes bibliographical references (p. [243]-256). Also available via the World Wide Web.

Orientador: Helena Redigolo Pezza / Banca: Edilene Cristina Ferreira / Banca: Danilo Luiz Flumignan / Banca: Aristeu Gomes Tininis / Banca: Eny Maria Vieira / Resumo: O presente trabalho tem por objetivo apresentar as metodologias desenvolvidas para a determinação de taninos hidrolisáveis, teor de compostos antioxidantes e conteúdo de aminoácidos totais em mel, café e suplementos de academia, respectivamente, empregando reflectância difusa, infravermelho próximo e análise por imagens como técnicas analíticas. Os métodos reflectométricos foram baseados em reações clássicas descritas na literatura, realizadas utilizando papel de filtro com barreiras hidrofóbicas como suporte sólido. Os modelos para determinação de compostos antioxidantes em amostras de café foram construídos a partir de medidas realizadas em espectrofotômetro NIR, aplicando modelos quimiométricos para quantificação. Para a determinação de aminoácidos totais em suplementos de academia, a determinação foi realizada utilizando duas técnicas: reflectância difusa e análise por imagens, a primeira seguiu o mesmo procedimento que os demais métodos reflectométricos, com o adicional do uso de um dispositivo USB, utilizado como fonte de aquecimento para a promoção da reação entre o analito e o reagente cromogênico, e a segunda baseou-se na análise de fotografias tiradas a partir de um aparelho celular com flash automático, e no uso do software ImageJ para a obtenção dos valores de absorbância. Em todos os métodos desenvolvidos foram realizados planejamentos experimentais para a otimização das variáveis. A determinação de taninos hidrolisáveis em amostras de mel foi realizada através d... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: This study aims to present methodologies developed for the determination of hydrolysable tannin content of antioxidant compounds and content of total amino acids in honey, coffee and health supplements, in that order, using diffuse reflectance near infrared and analysis images as analytical techniques. The reflectométricos methods are based on classical reactions described in the literature conducted using filter paper as barriers hydrophobic solid support. The method for the determination of antioxidant compounds in coffee samples was constructed from measurements NIR spectrophotometer, applying chemometric models for quantification. For the determination of total amino acids in health supplements, the determination was carried out using two techniques: diffuse reflectance and analysis of images, the first followed the same procedure as the other reflectometric methods and the second was based on analysis of photographs taken from a mobile device with automatic flash, by using the ImageJ software for obtaining absorbance data. For all the methods developed experimental design for the optimization of the variables were performed. Determination of hydrolyzable tannins in honey samples was performed by reaction between the analyte (hydrolyzable tannin) and KIO3 chromogenic reagent (potassium iodate), among acetate buffer (pH 4.75) in barrier filter paper hydrophobic, forming a colorful measurable product by diffuse reflectance at 525 nm. The experimental design provided as great volume for each reagent 10 L and concentrations optimized 3.1 x 104 mg L-1 KIO3 and pH 4.75. The Limits of detection (LOD) and quantification (LOQ) were obtained respectively 12,2 and 40,8 mg L-1. The determination of antioxidant compounds in coffee was performed by quantification of phenolic and flavonoids, as well as direct determination of total antioxidant capacity using infrared spectra close (NIR)... / Doutor

Submitted by JOÃO PAIXÃO DOS SANTOS NETO (joaopaixaoneto@gmail.com) on 2018-11-12T17:32:27Z No. of bitstreams: 1 Tese Definita João.pdf: 2906923 bytes, checksum: 845ed5e6f32d28d2764507fa3867c277 (MD5) / Rejected by Neli Silvia Pereira null (nelisps@fcav.unesp.br), reason: Solicitamos que realize correções na submissão seguindo as orientações abaixo: 1- O agradecimento à CAPES deve ter a seguinte redação: O presente trabalho foi realizado com apoio da Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Código de Financiamento 001 2- Falta o abstract (resumo em inglês) no repositório. Agradecemos a compreensão. on 2018-11-13T13:21:55Z (GMT) / Submitted by João Paixão dos Santos Neto (joaopaixaoneto@gmail.com) on 2018-11-13T13:45:38Z No. of bitstreams: 1 Tese Definita João.pdf: 2907242 bytes, checksum: c61b7c2515e87708154877088732effc (MD5) / Approved for entry into archive by Neli Silvia Pereira null (nelisps@fcav.unesp.br) on 2018-11-14T16:00:50Z (GMT) No. of bitstreams: 2 Tese Definita João.pdf: 2907242 bytes, checksum: c61b7c2515e87708154877088732effc (MD5) santosneto_jp_dr_jabo.pdf: 2907242 bytes, checksum: c61b7c2515e87708154877088732effc (MD5) / Made available in DSpace on 2018-11-14T16:00:50Z (GMT). No. of bitstreams: 2 Tese Definita João.pdf: 2907242 bytes, checksum: c61b7c2515e87708154877088732effc (MD5) santosneto_jp_dr_jabo.pdf: 2907242 bytes, checksum: c61b7c2515e87708154877088732effc (MD5) Previous issue date: 2018-10-10 / Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) / O objetivo geral deste trabalho foi conservar a qualidade de mangas ‘Palmer’ exportadas para a Europa e por objetivos específicos: I. estabelecer o melhor estádio de maturação para o transporte marítimo refrigerado e sob condições de atmosfera controlada (CA) determinado usando técnicas não-destrutivas de espectroscopia de infravermelho próximo (NIRS); II. desenvolver modelos de calibração e predição de matéria-seca (MS) e sólidos solúveis (SSC); III. comparar se a abordagem proposta difere do manuseio tradicional de mangas em uma remessa de mangas por frete marítimo para a Europa. No trabalho sobre a determinação dos índices de maturidade de manga ‘Palmer’ utilizando espectrômetro portátil de infravermelho próximo (VIS-NIR) o melhor modelo de calibração para SSC foi obtido usando espectros pré-processados com variação normal padrão (SNV), primeira derivada de Savitizk-Golay e janela de 699-999 nm. Para os SSC, foi observado um RMSECV de 1,39 % com um R2CV de 0,87 e RPD de 2,77. Melhores resultados foram observados para o modelo de calibração de MS construído com espectros em bruto usando a janela de 699-981 nm (RMSECV de 8,81 g.kg-1, R2CV de 0,84 e RPD de 2,51) e os modelos de calibração para firmeza foram considerados ruins. No estudo sobre armazenamento refrigerado de mangas ‘Palmer’ classificadas com base no teor de MS usando espectrômetro portátil de infravermelho próximo (VIS-NIR) foi possível concluir que o modelo PLS desenvolvido com as mangas da safra 2015/2016 não foi adequado para predizer o teor de MS em frutos da safra 2016/2017 (R² não ajustado). Assim, os espectros VIS-NIR da safra 2016/2017 foram incorporados ao conjunto de dados e um novo modelo foi desenvolvido (RMSEcv de 10,5 g.kg-1, R²P de 0,75). Com o novo modelo, as mangas ‘Palmer’ foram classificadas em dois estádios de maturação (150 g.kg-1 e 110 g.kg-1), o que resultou em diferenças de qualidade principalmente em relação à MS e SSC. No experimento sobre a resposta pós-colheita de mangas selecionadas de forma não-destrutiva quanto ao teor de MS durante e após o armazenamento sob atmosfera controlada (AC) foi observado que o modelo de regressão de mínimos quadrados parciais (PLSR) desenvolvido com frutos das safras de 2015/2016 e 2016/2017 de Cândido Rodrigues - SP, foi capaz de prever adequadamente o teor de MS de mangas produzidas em Petrolina - PE, mas com erro médio quadrático de predição bastante alto (RMSEP = 20,2 g.kg-1) e baixo R2P (0,19). Os espectros Vis-NIR de mangas produzidas em Petrolina - PE foram incorporados ao conjunto de dados e um novo modelo foi desenvolvido (RMSEv = 13,8 g.kg-1 e R²V = 0,63). Com o novo modelo PLSR foi possível selecionar mangas produzidas em Petrolina - PE com 150 g.kg-1 de MS, confirmando a robustez do modelo. Não foram observadas diferenças significativas na qualidade entre os frutos classificados com base na MS (150 g.kg-1) em relação às selecionadas usando a aparência visual (manuseio tradicional) durante o armazenamento sob AC (5 kPa O2 e 5 kPa CO2) por 30 dias a 14,2±2,3 °C e 92,6±2,8 % UR. No entanto, as mangas classificadas com 150 g.kg-1 de MS apresentaram menor desvio padrão, indicando um lote de frutos mais homogêneo. De modo geral, foi possível conservar a qualidade de mangas ‘Palmer’ a serem exportadas para a Europa, principalmente por selecionar frutos no estádio de maturação mais apropriado (150 g.kg-1 MS), resultando em lotes de frutos mais homogêneos. / The main objective of this study was to preserve the quality of ‘Palmer’ mangoes exported for the supply chain in Europe, and the specific objectives were: I. establish the best maturity stage for sea freight transportation and under controlled atmosphere (CA) conditions, determined using non-destructive of techniques near infrared spectroscopy; II. develop calibration and prediction models for dry matter (DM) and soluble solids content (SSC), III. compare if the proposed approach differs from the traditional mango handling to sea freight to Europe. In the study about the determination of the ‘Palmer’ mangoes maturity stages using near portable infrared spectrometer (VIS-NIR), the best calibration model for SSC was obtained using NIR spectra pre-processed with standard normal variation (SNV), first derivative of Savitizk-Golay, and 699-999 nm window. For SSC, it was observed a RMSECV of 1.39 %, with a R2CV of 0.87, and a RPD of 2.77. Better results could be seen for the DM calibration model built with raw spectra using the 699-981 nm window (RMSECV of 8.81 g.kg-1, R2CV of 0.84 and RPD of 2.51), and poor calibration models were obtained for firmness. In the study about refrigerated storage of ‘Palmer’ mangos sorted based on their DM content using near portable infrared spectrometer (VIS-NIR), it could be seen that the PLS model developed during 2015/2016 mango season was not appropriate to predict the DM content of fruit from 2016/2017 season (R² not adjusted). Therefore, the VIS-NIR spectra from 2016/2017 season were incorporated into the dataset and a new model was developed (RMSEcv of 10.5 g.kg-1, R²P of 0.75). With the new model, ‘Palmer’ mangoes were sorted in two maturity stages (150 g.kg-1 and 110 g.kg-1), which resulted in quality differences, mainly in relation to DM and SSC. In the study about the postharvest responses of mangos non-destructively sorted based on DM content during and after the storage under controlled atmosphere (CA), it was observed that the partial least squares regression (PLSR) model developed with fruit from 2015/2016 and 2016/2017 from Candido Rodrigues - SP, was able to properly predict the DM content of mangoes produced in Petrolina - PE, but with a high root mean square error of prediction (RMSEP = 20.2 g kg-1) and low R2P (0.19). Therefore the VIS-NIR spectra from mangoes produced in Petrolina – PE were incorporated into the dataset and a new model was developed (RMSEv of 13.8 g kg-1 and R²V of 0.63). With the new PLSR model, it was possible to sort mangoes produced in Petrolina – PE with 150 g.kg-1 DM, confirming the robustness of the model. It was not observed significant quality differences between fruit sorted based on DM (150 g.kg-1) in relation to visual appearance sorted fruit (traditional handling) during CA storage (5 kPa O2 and 5 kPa CO2) for up to 30 days at 14.2±2.3 °C and 92.6±2.8 % RH. However, the mangoes sorted with 150 g.kg-1 DM showed lower standard deviation, indicating a more homogeneous fruit batch. All in all, it was possible to preserve the quality of ‘Palmer’ mango to be exported to Europe, mainly by sorting the fruit in the appropriate maturity stage (150 g.kg-1 DM) with a more homogeneous fruit batch. / 2015/25631-9