Understanding complex analytical data by a supervised correlation coefficient obtained from random forest.

Area of Science:

• Chemometrics
• Analytical Chemistry
• Bioinformatics

Background:

• Multivariate chemometric methods are crucial for analyzing complex analytical data from techniques like mass spectrometry and NMR.
• Interpreting variables contributing to predictions is as important as accurate sample classification or quantification.
• Existing variable selection algorithms, such as partial least squares regression and random forest, have limitations in fully capturing variable interdependencies.

Purpose of the Study:

• To illustrate the application of Surrogate Minimal Depth (SMD) for variable selection and relationship analysis in complex analytical data.
• To demonstrate how SMD incorporates variable interdependencies into variable importance assessment.
• To highlight the utility of SMD in interpreting predictive models derived from diverse analytical data types.

Main Methods:

• Utilizing surrogate variables derived from random forest models within the SMD framework.
• Quantifying the mutual impact of variables on predictive models using a supervised correlation coefficient.
• Applying SMD to various types of complex analytical data, including mass spectrometry, NMR, and vibrational spectroscopy.

Main Results:

• SMD effectively incorporates variable relationships into the analysis of variable importance.
• The method provides a supervised correlation coefficient to express the mutual impact of variables.
• Demonstrated successful application of SMD across different analytical data types, showcasing its versatility.

Conclusions:

• SMD offers a powerful approach for interpreting complex analytical data by elucidating variable relationships.
• The method enhances the understanding of how individual variables and their interactions influence predictive models.
• SMD holds significant potential for advancing chemometric analysis in various scientific domains.