Multidimensional scaling informed by F-statistic: Visualizing grouped microbiome data with inference.

Area of Science:

• Microbial Ecology
• Bioinformatics
• Computational Biology

Background:

• Multidimensional scaling (MDS) is crucial for dimensionality reduction in microbial ecology, preserving sample distances.
• Existing MDS methods often require prior assumptions, limiting their use in general microbiome analysis.

Purpose of the Study:

• Introduce F-informed MDS, a novel MDS-based ordination method.
• Address limitations of traditional MDS by integrating the F-statistic for data distribution configuration.
• Improve statistical rigor and applicability in microbiome data analysis.

Main Methods:

• Developed F-informed MDS, configuring data distribution using the F-statistic (ratio of dispersions).
• Utilized semisynthetic datasets for method evaluation.
• Assessed robustness to hyperparameter selection and statistical significance.

Main Results:

• F-informed MDS demonstrates robustness to hyperparameter choices.
• The method maintains statistical significance during ordination.
• Quality metrics confirm comparable performance to state-of-the-art methods in preserving data structures.
• Application to diatom-associated bacteria revealed host-response patterns.

Conclusions:

• F-informed MDS offers a statistically grounded refinement of MDS for microbial ecology.
• The method aligns ordination results with statistical tests, aiding broader multidimensional data analysis.
• This visualization tool can be integrated into standard microbiome analysis workflows.