Pareto

Pareto evaluates clustering algorithms by identifying optimal trade-offs among multiple supervised performance metrics using Pareto fronts for cytometry and single-cell sequencing datasets.

Key Features:

• Integrated Multi-Metric Evaluation: Leverages complementary perspectives from several supervised clustering performance metrics simultaneously to provide a holistic assessment of clustering algorithms.
• Pareto Fronts Methodology: Uses Pareto fronts to identify solutions that offer optimal trade-offs among multiple evaluation criteria and to rank algorithms by their balance across metrics.
• Systematic Parameter Sampling: Employs Latin Hypercube sampling to systematically explore algorithm parameter values and reduce confounding from (un)fortunate parameter selections.
• Comparative Study Implementation: Implements a comparative study evaluating ChronoClust, FlowSOM, and Phenograph across four common performance metrics and four cytometry benchmark datasets.

Scientific Applications:

• Algorithm selection for cytometry and single-cell sequencing: Assists researchers in identifying optimal clustering algorithms for cytometry and single-cell sequencing datasets by providing nuanced multi-metric comparisons that improve translatability beyond benchmark datasets.

Methodology:

Utilizes multiple supervised clustering performance metrics; applies Pareto fronts to determine optimal trade-offs between these metrics; and systematically samples algorithm parameters using Latin Hypercube sampling.