Extended Data Fig. 1: Effect of choice of cost matrix metric in OT and several booster options (platform normalization, multicenter, bisection) on annotation quality.

L2 error (y axis, top; relevant for mixture annotations) and fraction of objects where the maximum annotation disagrees with the reference (y axis, bottom; relevant for classification problems) for different methods (colors) on different use cases (x axis). Baseline TACCO used the Bhattacharyya metric, platform normalization, multicenter, and bisection. Single cell: annotation of mouse colon scRNA-seq with itself as reference; Mixture: Gaussian mixtures of mouse colon scRNA-seq (Fig. 2a); Dropout: simulated Dropout dataset (Fig. 3a); Ambient: simulated ambient dataset (Fig. 3b), Differentiation: hematopoiesis scRNA-seq (Fig. 3c).