def load_epi(average='geometric'):
import singlet
ds_dev = load_epi_development(average=False)
ds_adu = load_epi_TMS(average=False)
dsepi = singlet.concatenate([ds_dev, ds_adu], missing='pad')
if not average:
return dsepi
if average == 'geometric':
dsepi.counts.log(inplace=True)
dsepi_av = dsepi.average('samples', by=['cellSubtype', 'Timepoint'])
return dsepi_av
}, )
ds_tm.samplesheet['Tissue'] = ds_tm.samplesheet['tissue']
ds_tm.samplesheet['Mousename'] = 'TM' + ds_tm.samplesheet['mouse.id']
ds_tm.samplesheet['Dataset'] = 'TabulaMuris'
ds_tm.samplesheet['Timepoint'] = '3MO'
# They are all endos, without further classification (ouch!)
ds_tm.samplesheet['Cell Type'] = ds_tm.samplesheet['cell_ontology_class']
ds_tm.samplesheet['cellSubtype'] = 'unknown (TM)'
print('Restrict TM to lung')
ds_tm0 = ds_tm
ds_tm = ds_tm0.query_samples_by_metadata('Tissue == "Lung"')
print('Concatenate datasets')
dsme = concatenate([ds, ds_tm], missing='pad')
print('Normalize')
dsme.counts.normalize('counts_per_million', inplace=True)
if False:
print('Feature selection')
features = dsme.feature_selection.overdispersed_within_groups(
'Mousename', inplace=False)
dsf = dsme.query_features_by_name(features)
print('PCA')
dsc = dsf.dimensionality.pca(n_dims=50,
robust=False,
return_dataset='samples')
dataset={
'path': fns_loom['schyns'],
'index_samples': 'CellID',
'index_features': 'Gene',
'bit_precision': 32,
},
)
ds_sc.samplesheet['cellSubtype'] = ds_sc.samplesheet['ClusterName']
print('Merge the data by hand')
genes = np.intersect1d(ds.featurenames, ds_sc.featurenames)
ds.query_features_by_name(genes, inplace=True)
ds.samplesheet['Data source'] = 'Domingo-Gonzales\net al. (this paper)'
ds_sc.query_features_by_name(genes, inplace=True)
ds_sc.samplesheet['Data source'] = 'Schyns et al.'
dsme = concatenate([ds, ds_sc])
dsme.counts.normalize('counts_per_million', inplace=True)
print('Feature selection')
features = dsme.feature_selection.overdispersed_within_groups(
'Data source', inplace=False)
dsf = dsme.query_features_by_name(features)
print('PCA')
dsc = dsf.dimensionality.pca(n_dims=30, robust=False, return_dataset='samples')
print('Embedding')
vs = dsc.dimensionality.umap()
dsme.samplesheet['umap1'] = vs.iloc[:, 0]
dsme.samplesheet['umap2'] = vs.iloc[:, 1]
else:
cache = pd.read_csv(
'../../data/sequencing/me1/northstar_with_Palantir.tsv',
sep='\t',
index_col=0)
vs = cache.iloc[:, :2]
northstar_assignment = cache['northstar_assignment']
print('Make dataset with merged')
genes = np.intersect1d(ds.featurenames, dsP.featurenames)
ds.query_features_by_name(genes, inplace=True)
ds.samplesheet['Data source'] = 'new_data'
ds.samplesheet['Cell Subtype'] = 'ME1'
dsPsub.query_features_by_name(genes, inplace=True)
dsPsub.samplesheet['Data source'] = 'Palantir'
dsme = singlet.concatenate([dsPsub, ds])
dsme.samplesheet['northstar_assignment'] = northstar_assignment
new_clusters = [x for x in np.unique(ns.membership) if x.isdigit()]
print('Plot embedding')
genes = ['Data source', 'Cell Subtype', 'northstar_assignment']
cmaps = {
#'clusters': an.uns['cluster_colors'],
'Cell Subtype': {
'HSC': 'deeppink',
'Ery-precursor': 'lawngreen',
'Mono': 'red',
'Mono-precursor': 'purple',
'CLP': 'tan',
'pDC': 'lightseagreen',
'Ery': 'forestgreen',
ns.compute_pca()
ns.compute_similarity_graph()
print('Cluster graph')
ns.cluster_graph()
print('Compute embedding')
vs = ns.embed('umap')
print('Make dataset with merged')
genes = np.intersect1d(ds.featurenames, ds_tm.featurenames)
ds.query_features_by_name(genes, inplace=True)
ds.samplesheet['Data source'] = 'new_data'
ds_tm.query_features_by_name(genes, inplace=True)
ds_tm.samplesheet['Data source'] = 'Tabula Muris'
dsme = concatenate([ds_tm, ds])
dsme.samplesheet['northstar assignment'] = np.concatenate(
[ds_tm.samplesheet['Cell Subtype'].values, ns.membership], )
new_clusters = [x for x in np.unique(ns.membership) if x.isdigit()]
print('Plot embedding')
genes = ['Data source', 'Cell Subtype', 'northstar assignment', 'Tissue']
cmaps = {
'Data source': {
'Tabula Muris': 'darkred',
'new_data': 'steelblue'
},
'Cell Subtype': {
'Mac IV': 'darkolivegreen',
'Mac II': 'lime',
'Mac I': 'seagreen',
if True:
print('Cxcl12/Cxcr4/Cxcr7')
import singlet
dsa = []
for g1, csts1 in ct_groups.items():
if g1 == 'Epi':
continue
dsia = ds0.query_samples_by_metadata('cellSubtype in @csts1',
local_dict=locals()).average(
'samples',
by='TimepointHO')
dsia.obs['new_name'] = [(g1, tp) for tp in dsia.samplenames]
dsia.rename('samples', 'new_name', inplace=True)
dsa.append(dsia)
dsa = singlet.concatenate(dsa)
dsepia = dsepi.average('samples', by=['cellSubtype', 'Timepoint'])
genes = ['Cxcl12', 'Cxcr4', 'Ackr3']
df = dsa.counts.loc[genes]
df.columns = pd.MultiIndex.from_arrays(
[[x[0] for x in df.columns], [x[1] for x in df.columns]], )
for (cst, tp) in dsepia.samplenames:
if tp not in tps_epi:
continue
tmp = pd.Series(np.zeros(len(genes)), index=genes)
for gene in genes:
if gene in dsepia.featurenames:
tmp[gene] = dsepia.counts.loc[gene, (cst, tp)]