Selection of single cell clustering methodologies through rank aggregation of multiple performance measures

TL;DR

This paper introduces an ensemble ranking approach using multiple performance measures to objectively select single-cell clustering methods tailored to specific datasets.

Contribution

It adapts existing evaluation measures and employs aggregation schemes to improve the selection process of clustering methodologies for single-cell data.

Findings

Ensemble of validation measures improves clustering method selection.

Ranking based on dataset-specific preferences enhances objectivity.

Method demonstrates effectiveness across six datasets.

Abstract

As single-cell gene expression data analysis continues to grow, the need for reliable clustering methods has become increasingly important. The prevalence of heuristic means for method choice could lead to inaccurate reports if comprehensive evaluation of the methods is omitted. Typical comparisons of methods fail to address the complexity presented by the data, transformations, or internal parameters. Previous work in the field of microarray data provided measures to evaluate the stability characteristic of clustering algorithms. Additional work on aggregation in the same era presented a way to compare multiple methodologies using several performance measures. In this paper, we provide adaptations to these measures and employ two aggregation schemes to create ranked lists of method and parameter choices for six unique datasets. Our findings demonstrate that an ensemble of validation measures, combined with ranking based on measures' dataset specific preferences, provides an objective way to select clustering methodologies, taking into account characteristic evaluation from each measure.