Journal of Chemometrics最新文献

The relationship between the Mahalanobis distance and the χ² and Hotelling T² distributions is described. The methods are illustrated by two case studies, a 40 × 2 simulation and a 54 × 34 dataset consisting of the NMR of metabolic extracts from maize harvested at 8.5°C. It is shown that if the number of samples is close to but still greater than the number of variables, p values calculated using T² as usually defined are higher than should be normally expected. This is interpreted as a problem caused by the difficulty of determining the number of degrees of freedom in multivariate matrices. It is recommended to use the χ² distribution in preference when calculating multivariate p values for probabilities that a sample belongs to a predefined distribution. When the sample size is large relative to the number of variables, there is little point in using T², and when small, T² can predict misleading p values due to problems determining the number of degrees of freedom.

Gas chromatography-ion mobility spectrometry (GC-IMS) is a powerful analytical technique for characterizing volatile organic compounds (VOCs). However, variability in retention and drift times frequently complicates peak alignment and downstream analyses. To address this challenge, we introduce RD-PMA, a Rigid-Deformable Point-Matching Algorithm derived from and extending an established point-set registration framework. RD-PMA is specifically designed for GC-IMS data and incorporates dimension-dependent modeling, where rigid translation is applied in the drift-time dimension and nonlinear deformation is employed in the retention-time dimension. Robustness is further enhanced through an iterative outlier-removal procedure. Extensive evaluations across multiple GC-IMS datasets, including both homogeneous and heterogeneous conditions, as well as same-parameter and different-parameter scenarios, demonstrate that RD-PMA consistently outperforms conventional alignment approaches, such as internal reference-based shifting and other point-matching methods. The algorithm preserves structural consistency while delivering high alignment accuracy and resilience to noise and parameter variability. Collectively, these findings establish RD-PMA as a reliable and scalable solution for GC-IMS peak alignment, with broad applicability to large-scale VOC profiling in chemometrics, food science, and clinical research.

Alzheimer's disease (AD) involves multiple pathogenic pathways, yet current therapeutic strategies remain largely symptomatic and focused on single molecular targets, underscoring a critical existing gap, emphasizing the need for rational multitarget drug design. Addressing this limitation, the present study employed an integrated in silico framework to identify multitarget 1,3,4-oxadiazole derivatives against acetylcholinesterase (AChE), butyrylcholinesterase (BChE), and glycogen synthase kinase-3β (GSK3β). A dataset of 273 reported compounds was used to develop robust QSAR models via Genetic Algorithm-Multiple Linear Regression (GA-MLR), exhibiting strong internal (R² = 0.638–0.758, Q²_LOO = 0.609–0.736) and satisfactory external predictivity (Q²_F1 − Q²_F2 = 0.566–0.800). Subsequent virtual screening of 3404 1,3,4-oxadiazoles from BindingDB identified 72 promising hits (IC₅₀ ≤ 300 nM), which underwent molecular dynamics (MD) docking. MD-based CDOCKER analysis highlighted compound 2851 with superior binding energies and stable key interactions compared to donepezil, supported by binding free energy (ΔG) calculations. ADMET evaluation indicated favorable pharmacokinetic and toxicity profiles, while density functional theory (DFT) analysis revealed enhanced reactivity and lower band gaps. This integrated computational workflow identified compound 2851 as a promising MT therapeutic agent for AD.

We introduce Auditory Analytics, a methodological framework that utilizes data sonification for scientific discovery. Auditory Analytics describes a cycle of collecting and deriving datasets, mapping data to audible signals (sonification), analytical listening, hypothesis formulation, and tool building, where human insights from any stage of the cycle can feed back into further iterations of the cycle in the form of new datasets, alternative mappings, and new models of the original phenomenon. In Auditory Analytics, the remarkable capacity of the human auditory system to extract meaningful information from complex soundscapes across multiple timescales is repurposed for exploring, interpreting, and analyzing data. As an illustration of how Auditory Analytics can be used to uncover relationships and dynamics in physical systems, we describe an earlier study in which we applied this methodology to investigate state transition passages in a molecular dynamics simulation of a small protein. Auditory Analytics helped us identify distinct hydrogen-bonding patterns associated with different rates of transit between folded and unfolded states, leading to a deeper understanding of the process of protein folding. A single, isolated data mapping—whether visual, auditory, haptic, mathematical, or verbal—provides an incomplete picture of reality; by adding the Auditory Analytics cycle to our portfolio of data interpretation tools, we can build a more complete picture of physical phenomena.

Multisensory tools are beginning to reformat research and education in chemistry and many other fields. For example, translating infrared spectra into sound (sonification) can unveil molecular facts that the eye might miss. Tactile approaches are used with 3-D printed scientific data such as electrophoresis gels. These scientific advances are expanding the way we study data and are part of a broader area known as immersive analytics. While immersive analytics cover all the human senses for data immersion, this perspective focuses on data visualization in virtual reality (VR). By visualizing chemometric data information in VR, a human can use their extensively trained pattern recognition and problem-solving skills to make final analysis decisions. As an example, presented is a feasibility study for one-class classification with a focus on correcting samples machine learning (ML) identified as false positives (FPs) and false negatives (FNs) that typically reside in the gray zone (class fringe samples) to respective true negatives (TNs) and positives (TPs). Conversely, while not expected in a well-designed VR universe, it is possible that TP and TN prediction samples in the gray zone could be classified as respective FNs and FPs by decisions in VR. Results are presented for three datasets showing the feasibility of using VR for classification decisions. These datasets are clam contamination and two cancer detection situations. Some brief comments on the potential of using VR to identify local structure within a class are also provided using a quantitative structure–activity relationship (QSAR) dataset.

Peanuts are an important food ingredient, and quality adulteration may pose health risks to consumers. Therefore, a rapid, accurate, and effective method for detecting peanut adulteration should be developed. This paper proposes a peanut adulteration identification method based on an electronic nose (e-nose) and a feature complementary calculation network (FCC-Net). First, volatile organic compound gas data of peanuts with different adulteration ratios are collected using an e-nose system. Then, leveraging the cross-sensitivity and temporal dynamic characteristics of the e-nose data, a feature complementary calculation module (FCCM) is proposed to extract deep gas features. Finally, based on the FCCM, a lightweight FCC-Net is designed to identify peanuts with varying adulteration levels. Experimental results demonstrate that FCC-Net outperforms classical lightweight deep learning models and state-of-the-art gas classification methods in terms of accuracy (97.33%), precision (96.92%), and recall (97.61%), while maintaining extremely low parameters (0.0102 M) and computational cost (0.3700 M). The combination of the e-nose system and FCC-Net provides an efficient and lightweight solution for peanut quality inspection.

The adoption of spectroscopy as a process analytical technology (PAT) modality in the pharmaceutical industry and related sectors has enabled advanced monitoring and control of manufacturing processes. Most applications of spectroscopic PAT instruments are dependent on chemometric multivariate data analysis (MVDA) methods to extract the relevant process data from the spectral measurements. However, calibrating and maintaining conventional MVDA methods is often burdensome, as it requires extensive time, material, and financial costs to generate the necessary representative samples and corresponding reference data. This calibration burden can be a barrier to the adoption of spectroscopic PAT in the pharmaceutical industry. Within this article, a classification of MVDA methods referred to as “lean chemometrics” is proposed and formalized. Lean chemometrics are time-saving, material-sparing, and cost-cutting MVDA methods that reduce the calibration burden relative to conventional chemometric methods of choice for spectroscopic PAT. Categories of various MVDA methods that are classifiable as lean chemometric techniques and practical considerations for integration of these techniques with PAT in common pharmaceutical PAT applications are discussed. The intention of lean chemometrics is to raise awareness of solutions that minimize the challenge of calibration burden toward improving spectroscopic PAT adoption.

Hyperspectral imaging (HSI) has emerged as a promising technique for microplastic detection through analysis of reflectance variations across multiple wavelengths. Traditional approaches have focused primarily on isolated microplastic particles, requiring labor-intensive separation procedures impractical for routine monitoring. The challenge of detecting microplastics directly on food surfaces stems from spectral similarities between microplastics and food matrices, making differentiation difficult using conventional methods. Leveraging recent advances in machine learning, this study explores how artificial neural networks (ANN) and one-dimensional convolutional neural networks (1D-CNN) can identify subtle spectral differences to detect microplastic particles on seafood without isolation. We systematically evaluated model architectures, preprocessing techniques, and hyperparameter configurations to optimize detection performance using hyperspectral data from tilapia samples contaminated with polyethylene microspheres. Our findings demonstrate that 1D-CNN models trained on hyperspectral data without dimensionality reduction significantly outperform other approaches, achieving object-level detection F1 scores of 0.963 for 600-μm particles and 0.950 for 300-μm particles. This detection strategy represents a substantial improvement over traditional methods and highlights the potential of deep learning–based approaches for non-destructive, efficient microplastic detection in food safety applications.