Extended Data Fig. 5: Comparison of batch effect correction methods on data.
From: Unified mass imaging maps the lipidome of vertebrate development
a. Signal intensity distributions across three acquisitions, color corresponding to the acquisition. On the left is the raw MALDI MSI data. To the right, the same data corrected with ComBat and uMAIA with related quantile-quantile plots representing the transformation between raw and normalized intensities b. Line plot tracking the mean molecule (mz=804.50) intensity and variance (shaded area) in background and foreground modes across consecutive samples for raw and uMAIA-corrected data. c. Scatter plot of mean signal vs standard deviation of signal average across sections for background and foreground modes for raw and uMAIA-corrected data (right) d. Images of 3 peaks (mz=804.50, 516.30, 835.50) with spatially variable distributions for raw data and data corrected with ComBat, scArches and uMAIA. e. Distribution plots showing principal component coordinates of pixels across different acquisitions for raw data and normalized data. The principal components appear to shift as a result of the distortion of many individual features, the normalization rescues their consistency between acquisitions. f. Heatmaps representing the number, proportion and presence of pixel clusters across sections (’cluster count’, ’cluster proportion’ and ’cluster presence’) for raw and corrected data.
