Spatial Communication¶

Spatial communication is a point of interest for us for the Spatial Transcriptomics Society, and we would like to find the conduction process of spatial communication.

Here, we introduce two method integrated in omicverse named COMMOT and flowsig.

We made three improvements in integrating the COMMOT and flowsig algorithm in OmicVerse:

• We reduced the installation conflict of COMMOT and flowsig, user only need to update OmicVerse to the latest version.
• We optimized the visualization of COMMOT and flowsig and unified the data preprocessing process so that users don't need to struggle with different data processing flows.
• We have fixed some bugs that could occur during function.

If you found this tutorial helpful, please cite COMMOT, flowsig and OmicVerse:

• Cang, Z., Zhao, Y., Almet, A.A. et al. Screening cell–cell communication in spatial transcriptomics via collective optimal transport. Nat Methods 20, 218–228 (2023). https://doi.org/10.1038/s41592-022-01728-4
• Almet, A.A., Tsai, YC., Watanabe, M. et al. Inferring pattern-driving intercellular flows from single-cell and spatial transcriptomics. Nat Methods (2024). https://doi.org/10.1038/s41592-024-02380-w

Preprocess data¶

Here we present our re-analysis of 151676 sample of the dorsolateral prefrontal cortex (DLPFC) dataset. Maynard et al. has manually annotated DLPFC layers and white matter (WM) based on the morphological features and gene markers.

This tutorial demonstrates how to identify spatial domains on 10x Visium data using STAGATE. The processed data are available at https://github.com/LieberInstitute/spatialLIBD. We downloaded the manual annotation from the spatialLIBD package and provided at https://drive.google.com/drive/folders/10lhz5VY7YfvHrtV40MwaqLmWz56U9eBP?usp=sharing.

Communication Analysis with COMMOT¶

Spatial communication inference¶

We will use the CellChatDB ligand-receptor database here. Only the secreted signaling LR pairs will be used.

Jin, Suoqin, et al. “Inference and analysis of cell-cell communication using CellChat.” Nature communications 12.1 (2021): 1-20.

We then filter the LR pairs to keep only the pairs with both ligand and receptor expressed in at least 5% of the spots.

Now perform spatial communication inference for these 250 ligand-receptor pairs with a spatial distance limit of 500. CellChat database considers heteromeric units. The signaling results are stored as spot-by-spot matrices in the obsp slots. For example, the score for spot i signaling to spot j through the LR pair can be retrieved from adata.obsp['commot-cellchat-Wnt4-Fzd4_Lrp6'][i,j].

(Optional) We read the ground truth area of our spatial data

This step is not mandatory to run, in the tutorial, it's just to demonstrate the accuracy of our clustering effect, and in your own tasks, there is often no Ground_truth

Visualize the communication signal in spatial space¶

Determine the spatial direction of a signaling pathway, for example, the FGF pathway. The interpolated signaling directions for where the signals are sent by the spots and where the signals received by the spots are from are stored in adata.obsm['commot_sender_vf-cellchat-FGF'] and adata.obsm['commot_receiver_vf-cellchat-FGF'], respectively.

Taken together, our findings indicate that FGF signaling is crucial for cortical folding in gyrencephalic mammals and is a pivotal upstream regulator of the production of OSVZ progenitors. FGF signaling is the first signaling pathway found to regulate cortical folding.

Communication signal inference with flowsig¶

Construct GEMs¶

We now construct gene expression modules (GEMs) from the unnormalised count data. For ST data, we use NMF.

If you want to study the genes in a GEM, we provide the ov.externel.flowsig.ul.get_top_gem_genes function for getting the genes in a specific GEM.

Construct the flow expression matrices¶

We construct augmented flow expression matrices for each condition that measure three types of variables:

1. Intercellular signal inflow, i.e., how much of a signal did a cell receive. For ST data, signal inflow is constructed by summing the received signals for each significant ligand inferred by COMMOT.
2. GEMs, which encapsulate intracellular information processing. We define these as cellwise membership to the GEM.

Intercellular signal outflow, i.e., how much of a signal did a cell send. These are simply ligand gene expression. 3. The kay assumption of flowsig is that all intercellular information flows are directed from signal inflows to GEMs, from one GEM to another GEM, and from GEMs to signal outflows.

For spatial data, we use COMMOT output directly to construct signal inflow expression and do not need knowledge about TF databases.

For spatial data, we retain spatially informative variables, which we determine by calculating the Moran's I value for signal inflow and signal outflow variables. In case the spatial graph has not been calculated for this data yet, FlowSig will do so, meaning that we need to specify both the coordinate type, grid or generic, and in the case of the former, n_neighs, which in this case, is 8.

Flow expression variables are defined to be spatially informative if their Moran's I value is above a specified threshold.

Learn intercellular flows¶

For spatial data, where there are far fewer "control vs. perturbed" studies, we use the GSP method, which uses conditional independence testing and a greedy algorithm to learn the CPDAG containing directed arcs and undirected edges.

For spatial data, we cannot bootstrap by resampling across individual cells because we would lose the additional layer of correlation contained in the spatial data. Rather, we divide the tissue up into spatial "blocks" and resample within blocks. This is known as block bootstrapping.

To calculate the blocks, we used scikit-learn's k-means clustering method to generate 20 roughly equally sized spatial blocks.

We use these blocks to learn the spatial intercellular flows.

Partially validate intercellular flow network¶

Finally, we will remove any "false positive" edges. Noting that the CPDAG contains directed arcs and undirected arcs we do two things.

First, we remove directed arcs that are not oriented from signal inflow to GEM, GEM to GEM, or from GEM to signal outflow and for undirected edges, we reorient them so that they obey the previous directionalities.

Second, we will remove directed arcs whose bootstrapped frequencies are below a specified edge threshold as well as undirected edges whose total bootstrapped frequencies are below the same threshold. Because we did not have perturbation data, we specify a more stringent edge threshold.

Visualize the result of flowsig¶

We can construct the directed NetworkX DiGraph object from adjacency_validated_filtered.

Cell-specific GEM¶

The first thing we need to be concerned about is which GEM, exactly, is relevant to the cell I want to study. Here, we use dotplot to visualize the expression of GEM in different cell types.

Visualize the flowsig network¶

We fixed the network function provided by the author and provided a better visualization.