Contact maps per cell

A cell’s contacts become a contact map — a sparse matrix of contact counts between genomic bins. The maps of all cells are kept in one .scool (one cooler per cell), linked to the ChromData of the cells, not copied into it; a pseudo-bulk map (all cells, or one cell type: pseudo-bulk maps) is a linked .cool / .mcool. The ChromData then has cells (with their metadata and embeddings) and, until a structure is computed, no spots.

import uchrom.datasets as ds

cd = ds.load("kim2020_scihic")                  # sci-Hi-C: 1,931 cells, cell types, per-cell 500 kb maps
cd.uns["linked_scool"]["per_cell"]              # the link: path (relative to the store), bin size, assembly
one = cd.load_linked_scool(key="per_cell", cell=cd.cells.index[0])   # one cell's map, a cooler
path = cd.linked_scool_path("per_cell")         # the .scool itself, for tools that read files

Task	API
read contact pairs: .pairs, phased pairs (hickit / Dip-C)	uchrom.io.read_pairs, read_phased_pairs
link per-cell maps / a pseudo-bulk map to the cells	cd.link_scool(path, key=...), cd.link_cool(path, key=...)
read them back	cd.load_linked_scool(key=, cell=), cd.linked_scool_path(key, cells=), cd.linked_cool_path(key)
link RNA / ATAC measured in the same cells	cd.link_anndata, cd.link_mudata; cd.linked_adata
ship a dataset as one store (atlas)	python -m chromdata.embedded STORE: the maps embedded, partitioned by chromosome
spots of a section in the tissue (spatial Hi-C)	cd.set_cell_positions, cd.cell_positions, cd.link_spatialdata

In an atlas store the maps are embedded rather than linked; the same calls read them over HTTP, and linked_scool_path(key, cells=...) exports the cells asked for once, one read per chromosome. Analyses take the ChromData and the key of its maps — FastHigashi, the scHiCluster features, the directionality index — rather than file paths. Linking, embedding and reading maps on a real store: the .chromdata.zarr store tutorial.

API: chromdata, uchrom.io.