CCC: Decide on an output format

[rcannood]

Typically three types of output:

• Per-Spot Prediction Scores (SpatialDM/LIANA+/NICHES)
• Per slide - i.e. one summary of interactions for the whole slide/sample (COMMOT/MISTy/LIANA+)
• Per Source-Target - i.e. a summary for the whole slide but with respect to cell type annotations (e.g. Giotto/CellChatv2)

[rcannood]

@dbdimitrov Could you format the above in a format like this?

AnnData object
 obs: 'dose_uM', 'timepoint_hr', 'raw_cell_id', 'hashtag_id', 'well', 'container_format', 'row', 'col', 'plate_name', 'cell_id', 'cell_type', 'split', 'donor_id', 'sm_name'
 obsm: 'HTO_clr', 'X_pca', 'X_umap', 'protein_counts'
 layers: 'counts', 'normalized'

[dbdimitrov]

Typically three types of methods:
I) Individual cell/spot resolution

AnnData object with n_obs × n_vars = N × I
    layers: 'ccc_pred'

where N is observations (i.e. cells/spots) and I interactions

II)

AnnData object with n_obs × n_vars = N × M
    obs: 'cell_type'
    uns: 'ccc_pred'

where ccc_pred is a pd.Dataframe with the following columns: {'source_ct', 'receiver_ct', 'ligand', 'receptor'}

III) Interactions summarized for pairs of cell types / protein across the whole slide

AnnData object with n_obs × n_vars = N × M
    uns: 'ccc_pred'

where ccc_pred is a pd.Dataframe with the following columns: {'ligand', 'receptor'}

• For II and III, it just returns the AnnData with the gene/protein expression matrix? or we discard all together and simply a pd.DataFrame.