Self organising maps for visualising and modelling

Area of Science:

• Analytical Chemistry
• Data Science
• Chemometrics

Background:

• Modern analytical chemistry generates complex datasets from diverse fields like metabolomics and environmental science.
• Traditional methods like Principal Component Analysis (PCA) and Partial Least Squares (PLS) have limitations with non-linear data and outlier sensitivity.
• Advancements in computing power make sophisticated methods like Self-Organising Maps (SOMs) more accessible and practical.

Purpose of the Study:

• To introduce and motivate the use of Self-Organising Maps (SOMs) in analytical chemistry.
• To highlight the advantages of SOMs over traditional methods for complex, non-linear datasets.
• To demonstrate the application and visualization capabilities of SOMs through case studies.

Main Methods:

• Self-Organising Maps (SOMs) for dimensionality reduction and visualization.
• Comparison of SOMs with Principal Component Analysis (PCA) and Partial Least Squares (PLS).
• Visualization techniques including best matching units, hit histograms, unified distance matrices, and component planes.
• Supervised SOMs for classification, multifactor data analysis, and variable selection.

Main Results:

• SOMs provide intuitive visualization methods, effectively utilizing map space.
• SOMs are less dependent on least squares solutions, error normality, and are less influenced by outliers compared to PCA and PLS.
• Supervised SOMs demonstrate utility in classification and Quality Control applications.
• Case studies confirm SOMs' applicability to diverse analytical challenges.

Conclusions:

• Self-Organising Maps (SOMs) are a powerful and accessible tool for analyzing complex datasets in modern analytical chemistry.
• SOMs offer significant advantages over traditional methods, particularly for non-linear data and when dealing with outliers.
• The visualization and classification capabilities of SOMs, as shown in case studies, make them valuable for various applications including cultural heritage, environmental, metabolomic, and pharmaceutical analysis.