Understanding mass spectrometry images: complexity to clarity with machine learning

Area of Science:

• Computational Biology
• Spectroscopy
• Data Science

Background:

• Hyperspectral mass spectrometry imaging (MSI) generates complex datasets, demanding advanced analytical methods.
• Artificial intelligence (AI) and machine learning (ML) offer powerful tools for MSI data interpretation, especially for biological samples.
• Recent focus is on non-linear ML techniques applied within the last five years.

Purpose of the Study:

• To review recent non-linear machine learning techniques for hyperspectral MSI data.
• To introduce and compare self-organizing map (SOM), SOM-RPM, t-SNE, and UMAP.
• To highlight their application in creating "similarity maps" for biological sample analysis.

Main Methods:

• Review of non-linear machine learning algorithms: SOM, SOM-RPM, t-SNE, and UMAP.
• Focus on techniques generating "similarity maps" representing spectral similarity.
• Analysis of algorithm functionalities and applications in biological MSI research.

Main Results:

• These non-linear ML techniques effectively handle complex MSI data.
• Similarity maps provide intuitive visualizations of spectral relationships within pixels.
• SOM, SOM-RPM, t-SNE, and UMAP demonstrate utility in exploring biological MSI data.

Conclusions:

• Non-linear ML methods, especially those producing similarity maps, offer a promising paradigm for hyperspectral MSI data exploration.
• These techniques enhance the visualization and understanding of complex biological samples.
• Further application of these methods can advance MSI research in life sciences.