Scholarlink

<p>Near-infrared (NIR) spectroscopy, in combination with multivariate calibration, plays a central role in process analytical technology (PAT) for quantitative analysis. While detection and quantification limits (LOD/LOQ) are critical performance indicators, their estimation in multivariate models remains non-trivial—especially in complex systems such as phytopharmaceuticals. This study compares two multivariate LOD/LOQ estimation frameworks: the variance-leverage based mLOD/mLOQ (IUPAC-compliant, geometry-driven) and the net analyte signal (NAS)–based approach, which quantifies analyte-specific signals orthogonalized against the model background. Partial least squares regression (PLS-R) models were developed for vitexin and isovitexin in phytopharmaceutical samples of <i>Passiflora incarnata</i>. The study included four NIR spectrometers (benchtop and handheld) implementing various optical principles, as well as two sample conditions (milled, intact). LOD/LOQ estimates were systematically evaluated against the influence of latent variable structure, model complexity, and orthogonal signal correction (OSC), a spectral preprocessing technique that explicitly enhances the isolation of analyte-specific variance. The results demonstrate how mLOD and NAS-LOD respond to model geometry when applied to phytopharmaceutical matrices. Both frameworks explicitly depend on the latent structure constructed in the PLS-R model; however, NAS-LOD is highly sensitive to analyte signal alignment and dispersion across latent variables, while mLOD reflects variance-weighted geometric detectability. OSC pretreatment improved the&#xa0;analyte signal concentration in early latent variables and led to markedly reduced NAS-LOD values, while mLOD values showed substantial responsiveness to latent space compression in the models. The combined use of both frameworks, supported by latent space diagnostics (analyte variance per LV, projection of pure standards), provides a multi-layered evaluation of model transparency and internal structure. These findings demonstrate that multivariate LOD/LOQ estimation can reinforce PAT-oriented method development not only as a validation metric, but as a diagnostic tool that provides standardized measures for controlling the&#xa0;NIR analytical method.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Standardizing NIR spectroscopy for PAT in phytopharmaceutical applications: multivariate detection and quantification limits of vitexin and isovitexin

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月21日

查看全文
  • Krzysztof B. Beć,
  • Justyna Grabska,
  • Jan-Clemens Cremer,
  • Christian W. Huck

摘要

Near-infrared (NIR) spectroscopy, in combination with multivariate calibration, plays a central role in process analytical technology (PAT) for quantitative analysis. While detection and quantification limits (LOD/LOQ) are critical performance indicators, their estimation in multivariate models remains non-trivial—especially in complex systems such as phytopharmaceuticals. This study compares two multivariate LOD/LOQ estimation frameworks: the variance-leverage based mLOD/mLOQ (IUPAC-compliant, geometry-driven) and the net analyte signal (NAS)–based approach, which quantifies analyte-specific signals orthogonalized against the model background. Partial least squares regression (PLS-R) models were developed for vitexin and isovitexin in phytopharmaceutical samples of Passiflora incarnata. The study included four NIR spectrometers (benchtop and handheld) implementing various optical principles, as well as two sample conditions (milled, intact). LOD/LOQ estimates were systematically evaluated against the influence of latent variable structure, model complexity, and orthogonal signal correction (OSC), a spectral preprocessing technique that explicitly enhances the isolation of analyte-specific variance. The results demonstrate how mLOD and NAS-LOD respond to model geometry when applied to phytopharmaceutical matrices. Both frameworks explicitly depend on the latent structure constructed in the PLS-R model; however, NAS-LOD is highly sensitive to analyte signal alignment and dispersion across latent variables, while mLOD reflects variance-weighted geometric detectability. OSC pretreatment improved the analyte signal concentration in early latent variables and led to markedly reduced NAS-LOD values, while mLOD values showed substantial responsiveness to latent space compression in the models. The combined use of both frameworks, supported by latent space diagnostics (analyte variance per LV, projection of pure standards), provides a multi-layered evaluation of model transparency and internal structure. These findings demonstrate that multivariate LOD/LOQ estimation can reinforce PAT-oriented method development not only as a validation metric, but as a diagnostic tool that provides standardized measures for controlling the NIR analytical method.

Graphical abstract