Extended Data Fig. 6: Performance of Alternative Factorizations of the Batch-Effect Parameters in the Normalization Framework.

a. Intensity distributions of 10 different compounds in the 72 hpf zebrafish after normalization using 3 alternative models for batch-effect factors. Colors correspond to different sections. b. Images of principal component coordinates of Raw, uMAIA and the 3 alternative models (see Methods: ‘Batch effect characterization’). Black boxes indicate residual batch-effects present in the images. c. PC coordinates of dataset after correction by various models across sections after normalization using the alternative models. d. Low dimensional representation (UMAP embedding) of pixels for dataset after normalization by alternative models. Pixels are colored by the section from which they originate. See Fig. 4F. e. Heatmaps representing matrices counting the number of pixels belonging to a given pixel cluster across sections (’cluster count’), distribution of cluster proportion across sections (’row-normalized’) and presence of cluster within section (’>10 pixels in cluster’) within each section for raw and corrected data. Clustering was performed for each dataset independently. See Extended Data Fig. 5 for a comparison with other tested methods. f. Average Wasserstein distances between intensity distributions of neighboring sections over molecules (left) and top 10 principal components (right) for raw data, data corrected with uMAIA and alternative models of batch-effect factorization. Boxplots indicate interquartile range and median.