Inspecting CCC patterns from spatial transcriptomics¶

This tutorial illustrates how Tensor-cell2cell can be used to explore cell-cell communication in distinct regions of a tissue using spatial transcriptomics. Here each region of the tissue is treated as a separate context.

Setups¶

This notebook requires installing scanpy, squidpy, and liana.

Data¶

RNA-seq data¶

Similar to the tutorial of using LIANA and MISTy, we will use an ischemic 10X Visium spatial slide from Kuppe et al., 2022. It is a tissue sample obtained from a patient with myocardial infarction, specifically focusing on the ischemic zone of the heart tissue.

The slide provides spatially-resolved information about the cellular composition and gene expression patterns within the tissue.

Normalize data

Visualize spot clusters or niches. These niches were defined by clustering spots from their cellular composition deconvoluted by using cell2location (Kuppe et al., 2022).

Cellular composition obtained from cell2location

Defining spatial contexts in the tissue¶

For defining spatial contexts within a tissue using spatial transcriptomics we simply divide the tissue in different regions through a square grid of a specific number of bins in each of the axes.

In this case, we divide our tissue into a 5x5 grid. A new column will be create in the adata object, specifically adata.obs['grid_cell'], to assign each spot or cell a grid window.

Here, cells or spots in one grid window do not get to interact with those that are within other grid windows, ignoring that cells in the interfaces between grid windows could also interact. For accounting for those cases, more complex approaches could be employed, as for example a sliding window strategy for defining contexts. In that case, spatial contexts will present some overlap in the cells or spots that are within them.

For simplicity we only tried the grid approach, which could be replaced by the sliding windows method, but it would also require to account for the overlap between spatial contexts. If you are interested in trying out this approach, see cell2cell.spatial.neighborhood.create_sliding_windows() and cell2cell.spatial.neighborhood.add_sliding_window_info_to_adata().

Protein-Protein Interactions or Ligand-Receptor Pairs¶

In this case, we use a list of LR pairs published with CellChat (Jin et al. 2021, Nature Communications).

Remove bidirectionality in the list of ligand-receptor pairs. That is, remove repeated interactions where both interactions are the same but in different order:

From this list:

We will have:

Generate a dictionary with function info for each LR pairs. Keys are LIGAND_NAME^RECEPTOR_NAME and values are the function in the annotation column in the dataframe containing ligand-receptor pairs.

Metadata¶

Metadata for the single cells

Tensor-cell2cell Analysis¶

Organize data to create tensor

In this dataset, contexts correspond to grid binst,

First, generate a dictionary indicating what condition is associated to each sample

Sort contexts to have them all together by condition, but donors in the same order within each condition

Generate list of RNA-seq data for each context

Build 4D-Communication Tensor

how='outer' is used to keep all cell types and LR pairs that are across all contexts.

outer_fraction=1/4.is to consider cell types and LR pairs that are present in at least 1/4 of contexts.

complex_sep='&' is used to specify that the list of ligand-receptor pairs contains protein complexes and that subunits are separated by '&'. If the list does not have complexes, use complex_sep=None instead.

(# Contexts, # LR pairs, # Sender cell types, # Receiver cell types) contained in the tensor, after all the preprocessing

Generate a list containing metadata for each tensor order/dimension - Later used for coloring factor plots

Elbow analysis for selecting Rank for Tensor-Factorization¶

Uncomment this if you prefer to run the elbow analysis

Tensor decomposition¶

Here the decomposition is performed into 8 factors. For determining an optimal number run the elbow analysis above, and replace rank=8 by rank=tensor.rank.

Results¶

 Plot factors¶

Color palettes for each dimension

Plot factor loadings

Top-5 LR pairs in each factor¶

Export an excel with all loadings¶

Visualize CCC patterns in space¶

Each factor or CCC pattern contains loadings for the context dimension. These loadings could be mapped for each spot or cell depending on what spatial context (grid window) they belong to. Thus, we can visualize the importance of each context in each of the factors, and see how the CCC patterns behave in space.

This behavior in space is useful to understand what set of LR pairs are used by determinant sender-receiver spot pairs in each region of the tissue. Regions with higher scores, indicate that LR pairs of that factor are used more there by the senders and receivers with high loadings.

Then, using the LR pair loadings, we can also identify LR interactions that are key in each factor, and link this with the spatial region with higher scores in the same factor.