We found that our bridge integration strategy consistently maximized this similarity metric, demonstrating that our procedure most effectively matched cells in the same biological state across modalities (Fig. 3c and Supplementary Fig. 3a). Consistent with our previous results, we found that the strongest improvements were observed when mapping rare cell types including plasma cells and dendritic cells (Supplementary Fig. 3b). As our procedure is compatible with multiple integration techniques, we compared the performance of bridge integration when using either mnnCorrect39 or Seurat v319 for the final alignment step, and observed very similar results (Supplementary Figure 3a,b). We also computed additional metrics based on the cluster labels originally assigned based on the scRNA-seq measurements44. These included the classification AUC (which factors in the prediction score assigned by each method), the query KNN purity (which assesses the fraction of each query cell neighbors that receive the correct annotation), and the multiclass binary cross-entropy (a commonly used classification metric51). In all cases, we consistently found that bridge integration exhibited superior performance (Supplementary Table 1).
