Dimensional cophenetic integrity: a method for evaluation of dimensionality reduction in MSI

Area of Science:

• Computational biology
• Data science
• Biotechnology

Background:

• Mass spectrometry imaging (MSI) generates high-dimensional data, posing challenges for visualization and analysis.
• Dimensionality reduction algorithms are crucial for simplifying MSI data for tasks like feature selection and clustering.
• Evaluating the effectiveness of these algorithms in preserving data structure is essential.

Purpose of the Study:

• To develop a novel method for assessing the structure preservation capabilities of dimensionality reduction algorithms in MSI data.
• To determine if optimized dimensionality reduction can improve data visualization and analysis outcomes.

Main Methods:

• Developed Dimensionality Cophenetic Integrity (DCI), a novel evaluation method based on cophenetic distances of hierarchically clustered samples.
• Applied DCI to assess structure preservation in reduced MSI data.
• Utilized DCI as an objective criterion for Bayesian optimization of dimensionality reduction hyperparameters.

Main Results:

• DCI effectively measures structure and pattern preservation of dimensionality reduction algorithms.
• DCI results correlate with expected tissue segmentation and image quality in synthetic MSI data.
• Optimized dimensionality reduction, guided by DCI, significantly preserves cluster relationships compared to default settings.
• Optimal hyperparameters for dimensionality reduction were found outside typical ranges.

Conclusions:

• Dimensionality Cophenetic Integrity is a reliable metric for evaluating dimensionality reduction in MSI.
• Optimizing dimensionality reduction using DCI enhances the quality of data analysis and visualization.
• This approach facilitates more accurate interpretation of complex MSI datasets.