Multimodal data integration refers to the task of combining together two datasets of different modalities of measurements (e.g., single-cell RNA sequencing and single-cell ATAC sequencing) on different observations of the same biological system. Integrating such measurements allows us to analyze the interaction between the different modalities, without requiring an explicitly joint measurement like sci-CAR or CITE-seq.
Datasets should include matched measurements from two modalities, which are contained in adata and adata.obsm["mode2"]. The task is to align these two modalities as closely as possible, without using the known bijection between the datasets.
Methods should create joint matrices adata.obsm["aligned"] and adata.obsm["mode2_aligned"] which reside in a joint space.
Metrics should evaluate how well the cells which are known to be equivalent are aligned in the joint space.