INFLECT

INFLECT identifies optimal metacluster numbers for high-dimensional cytometry, particularly mass cytometry, by evaluating marker interquartile range and unimodality to improve clustering interpretability and reproducibility.

Key Features:

• Optimal Cluster Determination: Overclusters data using FlowSOM to generate a range of metacluster numbers and evaluates them using marker interquartile range (IQR) and unimodality checks.
• Clustering Success Metrics: Calculates the fraction of marker distributions within each metacluster that pass predefined IQR and unimodality checks and plots this fraction against metacluster number.
• Inflection Point Analysis: Identifies the inflection point on the fraction-versus-metaclusters plot to determine the point of diminishing returns and an optimal cluster count.
• Versatile Clustering Approaches: Supports ConsensusClusterPlus metaclustering and hierarchical clustering, with hierarchical clustering noted as less computationally demanding while producing comparable results.

Scientific Applications:

• Mass cytometry dataset evaluation: Applied to four publicly available mass cytometry datasets of varying sizes and marker dimensions to determine optimal metacluster counts.
• Revealing cellular heterogeneity: Frequently identifies more metaclusters than published results, revealing distinct cell types encompassed by labeled populations.

Methodology:

Overclustering with FlowSOM; generation of a range of metacluster numbers; evaluation of marker interquartile range and unimodality within each metacluster; computation of the fraction of markers passing these checks per metacluster; plotting fraction versus metacluster number and identifying the inflection point; metaclustering via ConsensusClusterPlus or hierarchical clustering.