Scholarlink

Background <p>Reliable chemometric models require representative and well-distributed calibration and validation datasets.</p> Objective <p>The current study introduces a novel framework integrating diverse sampling strategies with multivariate modeling for the simultaneous UV–Vis quantification of aztreonam (AZM) and meropenem (MPM).</p> Methods <p>Three sampling techniques, including Monte Carlo (MC), Latin Hypercube Sampling (LHS), and Sobol Sequences (SS), were systematically evaluated in combination with Partial Least Squares (PLS), Genetic Algorithm–Assisted PLS (GA-PLS), and Artificial Neural Networks (ANN). Response surface strategy was followed for each sampling technique to assess their space coverage and points’ distribution across the experimental domain. Moreover, a hybrid variable-selection approach, namely, Genetic Algorithm Information Complexity–Partial Least Squares (GA-ICOMP-PLS), was also introduced to optimize PLS model variables. Different validation techniques, including nested cross-validation, Y-randomization tests, and noise-profiling, have been followed to ensure the models’ reliability.</p> Results <p>The ANN model trained with the SS technique achieved the highest predictive accuracy, reducing RMSE by 2.6% for AZM and 39.9% for MPM compared to standard PLS. However, LHS-GA-PLS delivered low prediction errors (22.8% AZM and 5.7% MPM), while MC-PLS showed lower consistency due to non-uniform sampling. GA-ICOMP-PLS further improved prediction performance for both analytes, with error reduction ranging from 35.1% to 63.6% compared to conventional PLS. Rigorous validation testing confirmed unbiased predictions and model resilience under realistic analytical conditions. </p> Significance <p>This is the first study to integrate structured sampling with chemometric modeling for β-lactam analysis. The approach improved model robustness, generalization, and sustainability, achieving high green metrics of SDS (Safety Data Sheets), AGREE (Analytical Greenness metric), and BAGI (Balanced Analytical Greenness Index) values of 0.84 and 85.0, respectively. The developed method represents a reliable, efficient, and eco-friendly strategy for pharmaceutical quality control and environmental monitoring.</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

Hybrid sampling chemometric strategy for accurate and sustainable quantification of co-administered β-lactam antibiotics

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月24日

查看全文
  • Shymaa S. Soliman,
  • Mona A. Abdel Rahman

摘要

Background

Reliable chemometric models require representative and well-distributed calibration and validation datasets.

Objective

The current study introduces a novel framework integrating diverse sampling strategies with multivariate modeling for the simultaneous UV–Vis quantification of aztreonam (AZM) and meropenem (MPM).

Methods

Three sampling techniques, including Monte Carlo (MC), Latin Hypercube Sampling (LHS), and Sobol Sequences (SS), were systematically evaluated in combination with Partial Least Squares (PLS), Genetic Algorithm–Assisted PLS (GA-PLS), and Artificial Neural Networks (ANN). Response surface strategy was followed for each sampling technique to assess their space coverage and points’ distribution across the experimental domain. Moreover, a hybrid variable-selection approach, namely, Genetic Algorithm Information Complexity–Partial Least Squares (GA-ICOMP-PLS), was also introduced to optimize PLS model variables. Different validation techniques, including nested cross-validation, Y-randomization tests, and noise-profiling, have been followed to ensure the models’ reliability.

Results

The ANN model trained with the SS technique achieved the highest predictive accuracy, reducing RMSE by 2.6% for AZM and 39.9% for MPM compared to standard PLS. However, LHS-GA-PLS delivered low prediction errors (22.8% AZM and 5.7% MPM), while MC-PLS showed lower consistency due to non-uniform sampling. GA-ICOMP-PLS further improved prediction performance for both analytes, with error reduction ranging from 35.1% to 63.6% compared to conventional PLS. Rigorous validation testing confirmed unbiased predictions and model resilience under realistic analytical conditions.

Significance

This is the first study to integrate structured sampling with chemometric modeling for β-lactam analysis. The approach improved model robustness, generalization, and sustainability, achieving high green metrics of SDS (Safety Data Sheets), AGREE (Analytical Greenness metric), and BAGI (Balanced Analytical Greenness Index) values of 0.84 and 85.0, respectively. The developed method represents a reliable, efficient, and eco-friendly strategy for pharmaceutical quality control and environmental monitoring.

Graphical Abstract