Pub Date : 2025-10-21DOI: 10.1016/j.chemolab.2025.105552
Ivana Hradecká , Kateřina Svobodová , Aleš Vráblík , Vladimír Hönig
This study compares the performance of near-infrared (NIR) and Raman spectroscopy in the quantitative analysis of used cooking oil (UCO), focusing on critical parameters such as acid value, density, and kinematic viscosity. Monitoring these properties ensures that the feedstock meets the necessary specifications for optimal biofuel production, contributing to the sustainability and performance of the final product. NIR and Raman spectroscopy offers significant advantages by enabling rapid, real-time and non-destructive measurements of several properties at once.
Partial least squares (PLS) was employed, enabling the correlation between reference results and spectral information obtained by NIR and Raman spectroscopy. NIR spectroscopy demonstrated superior performance compared to Raman spectroscopy in analyzing UCO properties. Results revealed the better performance of NIR spectroscopy for the measurement of acid value (R2P = 0.99, RMSEP = 0.087 mg KOH g⁻¹, RPD = 8.12), and kinematic viscosity at 40 °C (R2P = 0.97, RMSEP = 0.325 mm²/s, and RPD = 5.20). Raman spectroscopy was pointed out as the most suitable for the determination of density at 15 °C (R2P = 0.97, RMSEP = 0.167 kg m⁻³, RPD = 4.20). However, both techniques presented excellent results and are suitable for the accurate determination of UCO propreties.
{"title":"Comparison of NIR and Raman spectroscopy for determining used cooking oil properties using chemometric methods","authors":"Ivana Hradecká , Kateřina Svobodová , Aleš Vráblík , Vladimír Hönig","doi":"10.1016/j.chemolab.2025.105552","DOIUrl":"10.1016/j.chemolab.2025.105552","url":null,"abstract":"<div><div>This study compares the performance of near-infrared (NIR) and Raman spectroscopy in the quantitative analysis of used cooking oil (UCO), focusing on critical parameters such as acid value, density, and kinematic viscosity. Monitoring these properties ensures that the feedstock meets the necessary specifications for optimal biofuel production, contributing to the sustainability and performance of the final product. NIR and Raman spectroscopy offers significant advantages by enabling rapid, real-time and non-destructive measurements of several properties at once.</div><div>Partial least squares (PLS) was employed, enabling the correlation between reference results and spectral information obtained by NIR and Raman spectroscopy. NIR spectroscopy demonstrated superior performance compared to Raman spectroscopy in analyzing UCO properties. Results revealed the better performance of NIR spectroscopy for the measurement of acid value (R<sup>2</sup><sub>P</sub> = 0.99, RMSEP = 0.087 mg KOH g⁻¹, RPD = 8.12), and kinematic viscosity at 40 °C (R<sup>2</sup><sub>P</sub> = 0.97, RMSEP = 0.325 mm²/s, and RPD = 5.20). Raman spectroscopy was pointed out as the most suitable for the determination of density at 15 °C (R<sup>2</sup><sub>P</sub> = 0.97, RMSEP = 0.167 kg m⁻³, RPD = 4.20). However, both techniques presented excellent results and are suitable for the accurate determination of UCO propreties.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"268 ","pages":"Article 105552"},"PeriodicalIF":3.8,"publicationDate":"2025-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145464511","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this work, a chemometric methodology based on graph topology descriptors and CatBoost regression is proposed for predicting the physicochemical properties of anticancer drugs. Molecular structures were encoded as graphs, and degree-based topological descriptors were derived to capture their complexity. These descriptors were used in the construction of regression models predicting boiling point, molar refractivity, and polarizability. The first statistical analysis with linear and cubic regression demonstrated that models of order higher than unity were able to take into account the non-linear dependence of descriptors vs. molecular properties. CatBoost regression model was used for improved predictability and better interpretability. This model exhibits a coefficient of determination of 0.997 for the prediction of boiling point and superior performance across all the other two properties, with average absolute errors lower than 2%. Of importance, we identified several graph descriptors as important predictors, which confirmed their chemometric relevance. The method may contribute with useful information as a complementary method to current machine learning-based models used for prediction of drug properties in chemoinformatics or pharmaceutical drug development, it integrates chemical graph theory with intelligent reasoning and modeling for a more fault tolerant and generalized 1 solution to drug property prediction.
{"title":"Chemometric modelling of anticancer drugs using CatBoost regression and graphical derivatives","authors":"Yingxuan Huang , Muhammad Farhan Hanif , Eiman Maqsood , Mudassar Rehman","doi":"10.1016/j.chemolab.2025.105551","DOIUrl":"10.1016/j.chemolab.2025.105551","url":null,"abstract":"<div><div>In this work, a chemometric methodology based on graph topology descriptors and CatBoost regression is proposed for predicting the physicochemical properties of anticancer drugs. Molecular structures were encoded as graphs, and degree-based topological descriptors were derived to capture their complexity. These descriptors were used in the construction of regression models predicting boiling point, molar refractivity, and polarizability. The first statistical analysis with linear and cubic regression demonstrated that models of order higher than unity were able to take into account the non-linear dependence of descriptors vs. molecular properties. CatBoost regression model was used for improved predictability and better interpretability. This model exhibits a coefficient of determination <span><math><mrow><mo>(</mo><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup><mo>)</mo></mrow></math></span> of 0.997 for the prediction of boiling point and superior performance across all the other two properties, with average absolute errors lower than 2%. Of importance, we identified several graph descriptors as important predictors, which confirmed their chemometric relevance. The method may contribute with useful information as a complementary method to current machine learning-based models used for prediction of drug properties in chemoinformatics or pharmaceutical drug development, it integrates chemical graph theory with intelligent reasoning and modeling for a more fault tolerant and generalized 1 solution to drug property prediction.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"267 ","pages":"Article 105551"},"PeriodicalIF":3.8,"publicationDate":"2025-10-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145358697","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-11DOI: 10.1016/j.chemolab.2025.105550
Changrui Xie, Xi Chen
Industrial processes often exhibit multimode characteristics due to factors like load variations, equipment changes, and feedstock fluctuations. This paper introduces a Dirichlet Process-based Twofold-Robust Mixture Regression Model (DPR2MRM) for multimode processes. As a Bayesian nonparametric model, it automatically determines the number of mixture components from observed data using Dirichlet process mixture techniques, avoiding underfitting and overfitting. The model employs a Student's-t mixture model for input space learning, leveraging its long-tail properties for robust mode identification. For each mode, a regression model is built to capture the relationship between inputs and outputs, incorporating Student's-t noise to ensure robustness against output space outliers. The optimal posteriors of the model parameters are inferenced within a full Bayesian framework, and an analytical posterior predictive distribution is derived. The effectiveness of the DPR2MRM is demonstrated through a numerical example and two industrial applications.
{"title":"Robust soft sensor development based on Dirichlet process mixture of regression model for multimode processes","authors":"Changrui Xie, Xi Chen","doi":"10.1016/j.chemolab.2025.105550","DOIUrl":"10.1016/j.chemolab.2025.105550","url":null,"abstract":"<div><div>Industrial processes often exhibit multimode characteristics due to factors like load variations, equipment changes, and feedstock fluctuations. This paper introduces a Dirichlet Process-based Twofold-Robust Mixture Regression Model (DPR<sup>2</sup>MRM) for multimode processes. As a Bayesian nonparametric model, it automatically determines the number of mixture components from observed data using Dirichlet process mixture techniques, avoiding underfitting and overfitting. The model employs a Student's-<em>t</em> mixture model for input space learning, leveraging its long-tail properties for robust mode identification. For each mode, a regression model is built to capture the relationship between inputs and outputs, incorporating Student's-<em>t</em> noise to ensure robustness against output space outliers. The optimal posteriors of the model parameters are inferenced within a full Bayesian framework, and an analytical posterior predictive distribution is derived. The effectiveness of the DPR<sup>2</sup>MRM is demonstrated through a numerical example and two industrial applications.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"267 ","pages":"Article 105550"},"PeriodicalIF":3.8,"publicationDate":"2025-10-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145325250","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-10DOI: 10.1016/j.chemolab.2025.105547
Qing Wei , Vincent Chan , Kexin Xie , Kam-Wah Tsui , Xinwei Deng
For statistical analysis for various real-data applications, the relationship between the response and predictor variables is often complicated with certain constraints from the domain knowledge. While some predictor variables can be more important than others, it is important to enable model estimation and variable selection simultaneously for quantifying nonlinear patterns in the data with domain-knowledge constraints. In this work, we propose a penalized single index model that allows incorporation of prior shape information into the nonlinear function and performs shrinkage and variable selection simultaneously. The proposed methods also incorporate the need for robust model estimation and an efficient computational algorithm. The performance of the proposed method is evaluated by both simulation study and real-data studies of fuel consumption and DNA methylation.
{"title":"Estimation of penalized single index models with exact shape constraints","authors":"Qing Wei , Vincent Chan , Kexin Xie , Kam-Wah Tsui , Xinwei Deng","doi":"10.1016/j.chemolab.2025.105547","DOIUrl":"10.1016/j.chemolab.2025.105547","url":null,"abstract":"<div><div>For statistical analysis for various real-data applications, the relationship between the response and predictor variables is often complicated with certain constraints from the domain knowledge. While some predictor variables can be more important than others, it is important to enable model estimation and variable selection simultaneously for quantifying nonlinear patterns in the data with domain-knowledge constraints. In this work, we propose a penalized single index model that allows incorporation of prior shape information into the nonlinear function and performs shrinkage and variable selection simultaneously. The proposed methods also incorporate the need for robust model estimation and an efficient computational algorithm. The performance of the proposed method is evaluated by both simulation study and real-data studies of fuel consumption and DNA methylation.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"267 ","pages":"Article 105547"},"PeriodicalIF":3.8,"publicationDate":"2025-10-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145262444","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-10DOI: 10.1016/j.chemolab.2025.105541
Nicolás Hernández , Yoonsun Choi , Tom Fearn
We propose a novel Bayesian optimization framework for interval selection in Partial Least Squares (PLS) regression. Unlike traditional iPLS variants that rely on fixed or grid-based intervals, our approach adaptively searches over the discrete space of interval positions of a pre-defined width using a Gaussian Process surrogate model and an acquisition function. This enables the selection of one or more informative spectral regions without exhaustive enumeration or manual tuning. Through synthetic and real-world spectroscopic datasets, we demonstrate that the proposed method consistently identifies chemically relevant intervals, reduces model complexity, and improves predictive accuracy compared to full-spectrum PLS and stepwise interval selection techniques. A Monte Carlo study further confirms the robustness and convergence of the algorithm across varying signal complexities and uncertainty levels. This flexible, data-efficient approach offers an interpretable and computationally scalable alternative for chemometric applications.
{"title":"Bayesian optimization for interval selection in PLS models","authors":"Nicolás Hernández , Yoonsun Choi , Tom Fearn","doi":"10.1016/j.chemolab.2025.105541","DOIUrl":"10.1016/j.chemolab.2025.105541","url":null,"abstract":"<div><div>We propose a novel Bayesian optimization framework for interval selection in Partial Least Squares (PLS) regression. Unlike traditional iPLS variants that rely on fixed or grid-based intervals, our approach adaptively searches over the discrete space of interval positions of a pre-defined width using a Gaussian Process surrogate model and an acquisition function. This enables the selection of one or more informative spectral regions without exhaustive enumeration or manual tuning. Through synthetic and real-world spectroscopic datasets, we demonstrate that the proposed method consistently identifies chemically relevant intervals, reduces model complexity, and improves predictive accuracy compared to full-spectrum PLS and stepwise interval selection techniques. A Monte Carlo study further confirms the robustness and convergence of the algorithm across varying signal complexities and uncertainty levels. This flexible, data-efficient approach offers an interpretable and computationally scalable alternative for chemometric applications.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"267 ","pages":"Article 105541"},"PeriodicalIF":3.8,"publicationDate":"2025-10-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145262445","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-09DOI: 10.1016/j.chemolab.2025.105549
Zhen Li, Jiang Zhang
Leather derived from different animal sources exhibits significant differences in both performance and value. Traditional leather identification methods suffer from subjectivity, inefficiency, and high costs, motivating the need for rapid, objective, and cost-effective alternatives. To achieve rapid and non-destructive classification of leather types, our study introduces a novel combination of Raman spectroscopy and a one-dimensional convolutional neural network (1D-CNN) enhanced with a self-attention mechanism to efficiently capture subtle spectral differences among leather types. A total of 1066 Raman spectra from cow, sheep, pig, and crocodile leathers were collected. Spectral data underwent smoothing, baseline correction, and normalization. Seven samples from each leather class were randomly assigned to the training set, while the remaining three samples per class were designated as an independent validation set. Data augmentation was performed by adding Gaussian noise and applying slight spectral shifts to simulate real-world variability, expanding the training set to 3,810 samples. The proposed 1D-CNN model incorporates a self-attention mechanism to extract key spectral features and is compared with machine learning models and 1D-CNN models that do not integrate attention mechanisms. Experimental results demonstrate that our method outperforms existing approaches. After incorporating the self-attention mechanism, the model maintained a high accuracy during cross-validation, while its average classification accuracy on the independent test set increased from 92.11 % to 96.28 %. This result demonstrates that the proposed approach achieves enhanced generalization performance under different data partitioning schemes. This efficient, non-destructive, and reliable method not only enables accurate leather species identification and luxury goods authentication, but also shows promise for broader material classification and quality control applications.
{"title":"High-accuracy leather species identification via Raman spectroscopy and attention-enhanced 1D-CNN","authors":"Zhen Li, Jiang Zhang","doi":"10.1016/j.chemolab.2025.105549","DOIUrl":"10.1016/j.chemolab.2025.105549","url":null,"abstract":"<div><div>Leather derived from different animal sources exhibits significant differences in both performance and value. Traditional leather identification methods suffer from subjectivity, inefficiency, and high costs, motivating the need for rapid, objective, and cost-effective alternatives. To achieve rapid and non-destructive classification of leather types, our study introduces a novel combination of Raman spectroscopy and a one-dimensional convolutional neural network (1D-CNN) enhanced with a self-attention mechanism to efficiently capture subtle spectral differences among leather types. A total of 1066 Raman spectra from cow, sheep, pig, and crocodile leathers were collected. Spectral data underwent smoothing, baseline correction, and normalization. Seven samples from each leather class were randomly assigned to the training set, while the remaining three samples per class were designated as an independent validation set. Data augmentation was performed by adding Gaussian noise and applying slight spectral shifts to simulate real-world variability, expanding the training set to 3,810 samples. The proposed 1D-CNN model incorporates a self-attention mechanism to extract key spectral features and is compared with machine learning models and 1D-CNN models that do not integrate attention mechanisms. Experimental results demonstrate that our method outperforms existing approaches. After incorporating the self-attention mechanism, the model maintained a high accuracy during cross-validation, while its average classification accuracy on the independent test set increased from 92.11 % to 96.28 %. This result demonstrates that the proposed approach achieves enhanced generalization performance under different data partitioning schemes. This efficient, non-destructive, and reliable method not only enables accurate leather species identification and luxury goods authentication, but also shows promise for broader material classification and quality control applications.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"267 ","pages":"Article 105549"},"PeriodicalIF":3.8,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145262452","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-08DOI: 10.1016/j.chemolab.2025.105543
Harun Uslu , Bihter Das , Huseyin Alperen Dagdogen , Yunus Santur , Seval Yılmaz , Ibrahim Turkoglu , Resul Das
The discovery of novel therapeutic molecules against the Human Immunodeficiency Virus (HIV) remains a critical research priority due to the persistent global impact of the disease. Traditional drug discovery processes are often time-consuming, costly, and limited in predictive capacity at early stages. In this study, we propose a three-stage AI-supported framework that integrates deep learning and molecular docking to accelerate candidate identification. First, a customized Autoencoder–Long Short-Term Memory (LSTM) model was employed to generate novel molecular structures consistent with key pharmacokinetic rules. Second, a Geometric Deep Learning (GDL) model was designed to evaluate interactions with major HIV-1 targets, including integrase, protease, and reverse transcriptase. Finally, In silico docking simulations assessed binding affinities and inhibition constants. The framework generated molecules that not only complied with pharmacokinetic and drug-likeness criteria (e.g., QED, ADME, SAScore) but also demonstrated favorable binding properties, particularly towards HIV-1 reverse transcriptase. These findings highlight the potential of the proposed approach to complement early-stage drug discovery and to contribute to the design of promising lead compounds for further experimental validation.
{"title":"Discovery of new anti-HIV candidate molecules with an AI-based multi-stage system approach using molecular docking and ADME predictions","authors":"Harun Uslu , Bihter Das , Huseyin Alperen Dagdogen , Yunus Santur , Seval Yılmaz , Ibrahim Turkoglu , Resul Das","doi":"10.1016/j.chemolab.2025.105543","DOIUrl":"10.1016/j.chemolab.2025.105543","url":null,"abstract":"<div><div>The discovery of novel therapeutic molecules against the Human Immunodeficiency Virus (HIV) remains a critical research priority due to the persistent global impact of the disease. Traditional drug discovery processes are often time-consuming, costly, and limited in predictive capacity at early stages. In this study, we propose a three-stage AI-supported framework that integrates deep learning and molecular docking to accelerate candidate identification. First, a customized Autoencoder–Long Short-Term Memory (LSTM) model was employed to generate novel molecular structures consistent with key pharmacokinetic rules. Second, a Geometric Deep Learning (GDL) model was designed to evaluate interactions with major HIV-1 targets, including integrase, protease, and reverse transcriptase. Finally, <em>In silico</em> docking simulations assessed binding affinities and inhibition constants. The framework generated molecules that not only complied with pharmacokinetic and drug-likeness criteria (e.g., QED, ADME, SAScore) but also demonstrated favorable binding properties, particularly towards HIV-1 reverse transcriptase. These findings highlight the potential of the proposed approach to complement early-stage drug discovery and to contribute to the design of promising lead compounds for further experimental validation.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"267 ","pages":"Article 105543"},"PeriodicalIF":3.8,"publicationDate":"2025-10-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145262448","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-08DOI: 10.1016/j.chemolab.2025.105548
Jaume Béjar-Grimalt , Ángel Sánchez-Illana , Guillermo Quintás , Hugh J. Byrne , David Pérez-Guaita
Infrared and Raman spectroscopy hold great promise for clinical applications. However, the inherent complexity of the associated spectral data necessitates the use of advanced machine learning techniques which, while powerful in extracting biological information, often operate as black-box models. Combined with the absence of standardized datasets, this hinders model optimization, interpretability, and the systematic benchmarking of the growing number of newly developed machine learning methods. To address this, we propose a simulation-based framework for generating fully synthetic spectral datasets using Monte Carlo approaches for benchmarking. The artificial datasets mimic a wide range of realistic scenarios, including overlapping spectral markers and non-discriminant features and can be adjusted to simulate the effect of different parameters, such as instrumental noise, number of interferences, and sample size. These spectra are simulated through the generation of Lorentzian bands across the mid-infrared range, without specific reference to experimental data or chemical structures. We used the proposed methodology to compare different spectral marker identification protocols in a partial least squares discriminant analysis (PLS-DA), showing that the orthogonal PLS-DA (OPLS-DA) approach, when combined with marker selection based on VIP scores or the regression vector, yielded higher sensitivity, specificity, and interpretability than standard PLS-DA using the same selection criteria. This framework was further used to benchmark the classification capabilities of commonly employed machine learning algorithms, incorporating both linear and non-linear markers reflective of compositional variations across the target classes. Key findings were validated using real infrared spectra from human blood serum and saliva collected in the frame of a clinical study. Overall, the proposed approach provides a versatile sandbox environment for the systematic evaluation of data analysis strategies in vibrational spectroscopy, that can help experimentalists to better interpret spectral markers or data analysts focused on benchmarking and validating new algorithms.
{"title":"Monte Carlo peaks: Simulated datasets to benchmark machine learning algorithms for clinical spectroscopy","authors":"Jaume Béjar-Grimalt , Ángel Sánchez-Illana , Guillermo Quintás , Hugh J. Byrne , David Pérez-Guaita","doi":"10.1016/j.chemolab.2025.105548","DOIUrl":"10.1016/j.chemolab.2025.105548","url":null,"abstract":"<div><div>Infrared and Raman spectroscopy hold great promise for clinical applications. However, the inherent complexity of the associated spectral data necessitates the use of advanced machine learning techniques which, while powerful in extracting biological information, often operate as <em>black-box</em> models. Combined with the absence of standardized datasets, this hinders model optimization, interpretability, and the systematic benchmarking of the growing number of newly developed machine learning methods. To address this, we propose a simulation-based framework for generating fully synthetic spectral datasets using Monte Carlo approaches for benchmarking. The artificial datasets mimic a wide range of realistic scenarios, including overlapping spectral markers and non-discriminant features and can be adjusted to simulate the effect of different parameters, such as instrumental noise, number of interferences, and sample size. These spectra are simulated through the generation of Lorentzian bands across the mid-infrared range, without specific reference to experimental data or chemical structures. We used the proposed methodology to compare different spectral marker identification protocols in a partial least squares discriminant analysis (PLS-DA), showing that the orthogonal PLS-DA (OPLS-DA) approach, when combined with marker selection based on VIP scores or the regression vector, yielded higher sensitivity, specificity, and interpretability than standard PLS-DA using the same selection criteria. This framework was further used to benchmark the classification capabilities of commonly employed machine learning algorithms, incorporating both linear and non-linear markers reflective of compositional variations across the target classes. Key findings were validated using real infrared spectra from human blood serum and saliva collected in the frame of a clinical study. Overall, the proposed approach provides a versatile sandbox environment for the systematic evaluation of data analysis strategies in vibrational spectroscopy, that can help experimentalists to better interpret spectral markers or data analysts focused on benchmarking and validating new algorithms.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"267 ","pages":"Article 105548"},"PeriodicalIF":3.8,"publicationDate":"2025-10-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145325251","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-03DOI: 10.1016/j.chemolab.2025.105546
Muhammad Ali , Nasir Abbas , Shabbir Ahmad , Tahir Mahmood , Muhammad Riaz
Early detection of shifts in process mean is crucial for maintaining product quality and operational integrity in chemical industries. This paper proposes a new cumulative sum control chart named the CMD chart, that leverages an auxiliary variable for robust and efficient monitoring. The CMD chart is designed through various parameters, with control limits calibrated to ensure a desired average run length when in control. Its performance is assessed using multiple run-length metrics, including average run length, standard deviation, expected average run length, extra quadratic loss, relative average run length, and performance comparison index. An R Shiny app is also developed to enhance usability, simplify calibration and evaluation for different parameter combinations. Through extensive simulation across a broad range of shifts, the CMD chart consistently outperformed existing charts in quickly detecting shifts while minimizing false alarms. A practical case study in a polymerization reactor further highlighted effectiveness of CMD chart, demonstrating earlier, more accurate, and frequent detections of subtle shifts compared to competing methods. Overall, the CMD chart proves to be a robust and high-performing tool for process monitoring, making it highly relevant for modern chemical-engineering applications.
{"title":"On designing robust and efficient CUSUM chart for mean monitoring: An application in chemical engineering for polymerization reactors","authors":"Muhammad Ali , Nasir Abbas , Shabbir Ahmad , Tahir Mahmood , Muhammad Riaz","doi":"10.1016/j.chemolab.2025.105546","DOIUrl":"10.1016/j.chemolab.2025.105546","url":null,"abstract":"<div><div>Early detection of shifts in process mean is crucial for maintaining product quality and operational integrity in chemical industries. This paper proposes a new cumulative sum control chart named the CMD chart, that leverages an auxiliary variable for robust and efficient monitoring. The CMD chart is designed through various parameters, with control limits calibrated to ensure a desired average run length when in control. Its performance is assessed using multiple run-length metrics, including average run length, standard deviation, expected average run length, extra quadratic loss, relative average run length, and performance comparison index. An R Shiny app is also developed to enhance usability, simplify calibration and evaluation for different parameter combinations. Through extensive simulation across a broad range of shifts, the CMD chart consistently outperformed existing charts in quickly detecting shifts while minimizing false alarms. A practical case study in a polymerization reactor further highlighted effectiveness of CMD chart, demonstrating earlier, more accurate, and frequent detections of subtle shifts compared to competing methods. Overall, the CMD chart proves to be a robust and high-performing tool for process monitoring, making it highly relevant for modern chemical-engineering applications.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"267 ","pages":"Article 105546"},"PeriodicalIF":3.8,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145262453","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-03DOI: 10.1016/j.chemolab.2025.105545
Bin Li , Eizo Taira , Tetsuya Inagaki
Near-infrared spectroscopy (NIRS) calibration transfer faces significant challenges when deploying models across multiple instruments from different manufacturers, particularly because the inherently low molar absorptivity makes spectral data highly sensitive to minor variations in optical setup. This study presents two enhanced calibration transfer methods (ICTWM1 and ICTWM2) operating within the PLS latent variable space, utilizing dimensionality reduction to preserve analytically relevant variance while reducing noise interference. ICTWM1 employs spectral space transformation (SST) to correct PLS component scores between different instruments, while ICTWM2 selectively corrects the regression coefficients of the principal components with the highest variability.
The methods were validated using wheat protein analysis (7 secondary instruments from manufacturers A and B) and industrial sugarcane Brix determination (8 secondary instruments across geographically distributed facilities). ICTWM1 demonstrated superior performance, achieving 79.3 % relative performance compared to the primary instrument model using only 10 standardization samples on the wheat dataset, with improved cross-instrument consistency (standard deviations of 6.9 %) compared to traditional methods (>15 %). The method exhibited no manufacturer-dependent performance bias and maintained consistent performance across sample sizes ranging from 10 to 110. Under severely constrained sugarcane dataset with only 5 training samples, both ICTWM1 and ICTWM2 achieved good performance with mean RMSEP values of 0.14°Bx and 0.15°Bx, respectively, outperforming traditional calibration transfer methods.
ICTWM1 demonstrates improved sample efficiency and cross-manufacturer robustness through optimized transformation within PLS subspace. These characteristics make it a practical method for industrial NIRS applications requiring reliable calibration transfer with minimal standardization samples.
近红外光谱（NIRS）校准转移在不同制造商的多台仪器上部署模型时面临重大挑战，特别是因为固有的低摩尔吸收率使得光谱数据对光学设置的微小变化高度敏感。本研究提出了在PLS潜变量空间内运行的两种增强的校准传递方法（ICTWM1和ICTWM2），利用降维来保留分析相关方差，同时降低噪声干扰。ICTWM1采用谱空间变换（spectral space transformation， SST）对不同仪器间PLS成分得分进行校正，而ICTWM2则对变异性最高的主成分回归系数进行选择性校正。使用小麦蛋白分析（来自A和B制造商的7台二级仪器）和工业甘蔗糖度测定（分布在不同地理位置的8台二级仪器）对方法进行了验证。ICTWM1表现出优异的性能，与仅使用小麦数据集上10个标准化样本的主要工具模型相比，其相对性能达到79.3%，与传统方法相比，其跨工具一致性（标准偏差为6.9%）有所提高（> 15%）。该方法没有表现出与制造商相关的性能偏差，并且在从10到110的样本量范围内保持一致的性能。在只有5个训练样本的严格约束甘蔗数据集下，ICTWM1和ICTWM2均取得了较好的性能，RMSEP均值分别为0.14°Bx和0.15°Bx，优于传统的校准转移方法。ICTWM1通过优化PLS子空间内的变换，提高了样本效率和跨厂商鲁棒性。这些特性使其成为工业近红外光谱应用的实用方法，需要用最小的标准化样品进行可靠的校准转移。
{"title":"Enhanced PLS subspace-based calibration transfer method for multiple spectrometers using small standardization sample sets","authors":"Bin Li , Eizo Taira , Tetsuya Inagaki","doi":"10.1016/j.chemolab.2025.105545","DOIUrl":"10.1016/j.chemolab.2025.105545","url":null,"abstract":"<div><div>Near-infrared spectroscopy (NIRS) calibration transfer faces significant challenges when deploying models across multiple instruments from different manufacturers, particularly because the inherently low molar absorptivity makes spectral data highly sensitive to minor variations in optical setup. This study presents two enhanced calibration transfer methods (ICTWM1 and ICTWM2) operating within the PLS latent variable space, utilizing dimensionality reduction to preserve analytically relevant variance while reducing noise interference. ICTWM1 employs spectral space transformation (SST) to correct PLS component scores between different instruments, while ICTWM2 selectively corrects the regression coefficients of the principal components with the highest variability.</div><div>The methods were validated using wheat protein analysis (7 secondary instruments from manufacturers A and B) and industrial sugarcane Brix determination (8 secondary instruments across geographically distributed facilities). ICTWM1 demonstrated superior performance, achieving 79.3 % relative performance compared to the primary instrument model using only 10 standardization samples on the wheat dataset, with improved cross-instrument consistency (standard deviations of 6.9 %) compared to traditional methods (>15 %). The method exhibited no manufacturer-dependent performance bias and maintained consistent performance across sample sizes ranging from 10 to 110. Under severely constrained sugarcane dataset with only 5 training samples, both ICTWM1 and ICTWM2 achieved good performance with mean RMSEP values of 0.14°Bx and 0.15°Bx, respectively, outperforming traditional calibration transfer methods.</div><div>ICTWM1 demonstrates improved sample efficiency and cross-manufacturer robustness through optimized transformation within PLS subspace. These characteristics make it a practical method for industrial NIRS applications requiring reliable calibration transfer with minimal standardization samples.</div></div>","PeriodicalId":9774,"journal":{"name":"Chemometrics and Intelligent Laboratory Systems","volume":"267 ","pages":"Article 105545"},"PeriodicalIF":3.8,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145262446","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号