Bias-aware integration of single cell data from multiple modalities and technologies
High throughput single-cell technologies such as flow/mass cytometry, and RNA/CITE/ATAC-seq each have distinct biases in their data distributions. As a result, mapping cells from one dataset to another has so far had poor resolution, limiting the extent to which populations discovered using one platform (e.g. rare antigen-specific immune cells idenfied by cytometry) can be matched to atlases of well-characterized reference populations.
This project uses large-scale training datasets to learn the signal distribution characteristics of each assay, and to infer near-optimal transforms that allow high-resolution mapping of data from one modality onto another.
