Simplification of cell type calling

[MeyerBender]

Cell type and subtype calling is currently a bit cumbersome and should be simplified. Current syntax:

# processing and recomputing quantification
ds_processed = ds.pp[channels].pp.threshold(quantiles).pp.apply(medfilt2d, kernel_size=3).pp.add_quantification().pp.transform_expression_matrix(method='arcsinh')

# predicting major cell types
ds_with_ct_predictions = ds_processed.la.predict_cell_types_argmax(ct_marker_dict, key='_intensity', overwrite_existing_labels=True)

# merging processed image with segmentation etc. 
# thresholding specific markers and adding the final labels

There are a couple of things that could be simplified here:

• add a channel argument to pp.threshold. This way, we can still threshold selected channels, and simply threshold non-listed channels at 0. This removes the step of having one thresholded and one non-thresholded object.
• a hierarchical, all-in-one approach to cell type and subtype classification. This could for example look like this:

# predicting major cell types
ct_marker_dict = {'B': 'PAX5', 'T': 'CD3', 'Myeloid': 'CD11b', 'Dendritic': 'CD11c', 'Granulo': 'CD15', 'Macro': 'CD68', 'Stroma PDPN': 'Podoplanin', 'Stroma CD31': 'CD31', 'Stroma CD34': 'CD34', 'Stroma CD90': 'CD90', 'NK': 'CD56'}
ds = ds.la.predict_cell_types_argmax(ct_marker_dict)

# binarization of markers
# current syntax: ds.pp[marker].pp.threshold(intensity=threshold).pp.add_quantification(func=sp.percentage_positive, key_added='_perc_pos_filtered').la.threshold_labels({marker: perc_pos}, layer_key='_perc_pos_filtered')
ds = ds.pp.binarize(markers, thresholds)

# calling of cell subtypes with a provided hierarchy
subtype_dict = {
    'B': {
        'subtypes': [
            {
                'name': 'B_prol',
                'markers': ['Ki67']
            }
        ]
    },
    'T': {
        'subtypes': [
            {
                'name': 'T_h',
                'markers': ['CD4']
            },
            {
                'name': 'T_tox',
                'markers': ['CD8'],
                'subtypes': [
                    {
                        'name': 'T_tox_naive',
                        'markers': ['CD45RA']
                    },
                    {
                        'name': 'T_tox_mem',
                        'markers': ['CD45RO']
                    }
                ]
            }
        ]
    }
}

ds = ds.la.predict_cell_subtypes(subtype_dict)

[MeyerBender]

Addressed in #99, published with v0.6.0