Quantifying common and distinct information in single-cell multimodal data with Tilted-CCA

Multimodal single-cell technologies profile multiple modalities for each cell simultaneously and enable a more thorough characterization of cell populations alongside investigations into cross-modality relationships. Existing dimension-reduction methods for multimodal data focus on capturing the "union of information," producing a lower-dimensional embedding that combines the information across modalities. While these tools are useful, we develop Tilted-CCA to quantify the "intersection and difference of information", that is, a decomposition of a paired multimodal dataset into common axes of variation that is shared between both modalities and distinct axes of variation that is found only in one modality. Through examples, we show that Tilted-CCA enables meaningful visualization and quantification of the cross-modal information overlap. We also demonstrate the application of Tilted-CCA to two specific types of analyses. First, for single-cell experiments that jointly profile the transcriptome and surface antibody markers, we show how to use Tilted-CCA to design the target antibody panel to best complement the transcriptome. Second, for single-cell multiome data that jointly profiles transcriptome and chromatin accessibility, we show how to use the common embedding given by Tilted-CCA to identify development-informative genes and distinguish between transient versus terminal cell types.