INFLECT: an R-package for cytometry cluster evaluation using marker modality.

Area of Science:

• Computational Biology
• Bioinformatics
• Data Science

Background:

• High-dimensional cytometry data analysis, particularly mass cytometry, lacks robust methods for evaluating unsupervised clustering results.
• Existing methods rely on indirect measures like dimensionality reduction agreement or manual cell classification, introducing ambiguity and reproducibility issues.
• Overclustering can lead to a loss of statistical power, while underclustering fails to capture true biological heterogeneity.

Purpose of the Study:

• To introduce INFLECT, an R-package designed to provide insight into clustering results and determine an optimal number of clusters for mass cytometry data.
• To establish a quantitative method for assessing the quality of unsupervised clustering in high-dimensional cytometry datasets.
• To balance the capture of cellular heterogeneity with statistical power in data analysis.

Main Methods:

• INFLECT intentionally overclusters mass cytometry data using FlowSOM to capture fine-grained phenotypic subsets.
• It evaluates a range of metacluster numbers using marker interquartile range and distribution unimodality checks.
• The fraction of unimodal marker distributions is plotted against the number of metaclusters, identifying an inflection point that signifies optimal clustering.

Main Results:

• INFLECT was applied to four public mass cytometry datasets, consistently identifying a plateau in the unimodality score with a dataset-dependent inflection point.
• Both ConsensusClusterPlus and hierarchical clustering were tested, with hierarchical clustering offering computational efficiency with similar results.
• Comparison with labeled data revealed that INFLECT identified a higher optimal number of metaclusters, uncovering underlying cellular heterogeneity within broader labels.

Conclusions:

• INFLECT addresses a critical need for assessing clustering results in high-dimensional cytometry analysis.
• The package monitors marker distribution unimodality and interquartile range to guide optimal cluster number selection.
• The identified inflection point represents an optimal trade-off between capturing cellular heterogeneity and maintaining statistical power in mass cytometry data analysis.