Qualitative evaluation of chromatographic data from quality control schemes using a support vector machine

Area of Science:

• Analytical Chemistry
• Forensic Science
• Machine Learning in Quality Assurance

Background:

• Qualitative evaluation of chromatographic data is crucial for external quality assurance in analytical laboratories.
• Assessing data homogeneity among human experts, especially for analytes near the limit of detection, presents challenges.
• Variability in expert judgment can impact the reliability of analytical results in fields like anti-doping.

Purpose of the Study:

• To examine the homogeneity of chromatographic data evaluation among human experts.
• To develop a Support Vector Machine (SVM) as an automated and more consistent alternative for data evaluation.
• To identify factors influencing the quality of chromatographic analytical data.

Main Methods:

• Utilized 105 ion chromatograms from anti-doping control laboratories.
• Employed nine independent experts to qualitatively score chromatograms (0-4) and objective chromatographic parameters (peak width, asymmetry, resolution, S/N ratio).
• Developed and validated an SVM using a seven-fold cross-validation approach, incorporating both qualitative expert scores and objective data.

Main Results:

• Significant variability was observed among human experts in evaluating ion chromatograms.
• An SVM was successfully developed, closely mimicking the reasoning of a homogeneous group of human experts.
• The SVM demonstrated the potential for a scoring system to rank laboratories based on data quality.

Conclusions:

• A Support Vector Machine (SVM) offers a more homogeneous and automated approach to the qualitative evaluation of chromatographic data compared to human experts.
• The developed SVM can aid in identifying factors affecting chromatographic data quality, such as analytical techniques and reporting adherence.
• This approach enhances the reliability and consistency of external quality assurance schemes in analytical chemistry.