Fig. 1: Resolving conflicting cluster identities from decision-level integration.
From: Decision level integration of unimodal and multimodal single cell data with scTriangulate
Overview of scTriangulate. Using multiple competing annotation sources (algorithm, modality, and resolution), different statistical stability metrics are computed for each cluster and cell. Each cell is assigned to the winning cluster label based on a computed importance score (Shapley or Rank) (Steps 1–5, Methods). The reassign score considers the extent to which cells within a cluster can be reclassified to their own centroid based on the nearest neighbor. The TF-IDF score corresponds to the statistical strength of the nth-ranked feature (gene, ADT, peak) in a cluster. The SCCAF score implements a prior reliability metric (Single Cell Clustering Assessment Framework) that leverages multivariate multi-class logistic regression.
