from archimedes.functions.dataflow import input_var, input_json, output_json, parseModelAttr, buildTargetPath
from archimedes.functions.magma import question, connect
from archimedes.functions.list import unique, flatten
from archimedes.functions.environment import project_name
pdat = input_json("project_data")[project_name]
selection_options = pdat['selection_options']
seq_target = parseModelAttr(pdat['seq_h5_counts_data'])
q_start = [seq_target['model'], ['::has', seq_target['attribute']], '::all']
magma = connect()
options = dict([[
key,
unique(
flatten(
question(magma,
q_start + buildTargetPath(selection_options[key], pdat))))
] for key in list(selection_options.keys())])
output_json(options, 'selection_options')
# then follow the link to 'biospecimen' (1:many; but note this link type-info is not actually needed!),
# = ['biospecimen_group', 'biospecimen']
# then from there to travel up to that model's parent, 'subject'.
# = ['biospecimen_group', 'biospecimen', 'subject']
'experiment': ['sample', 'patient', 'experiment'],
'patient': ['sample', 'patient'],
'sample': ['sample'],
'sc_rna_seq': []
},
'color_options': {
# Format = <Label for the color-by drop down>: '<model>#<attribute>'
# Cluster, Tube, and Gene are standard options that do not need to be added here!
'Indication': 'experiment#name',
'Tissue': 'sample#tissue_type',
'Compartment': 'sc_rna_seq#biospecimen',
'Chemistry': 'sc_rna_seq#chemistry',
'Frozen tissue?': 'patient#ffpe_frozen'
# age, sex at birth, age at diag, bmi, smoker status, alcohol use, race, ethnicity, *past medical history (up to 10 per), time on ice
},
'selection_options': {
# Format = <Label of this selection item>: '<model>#<attribute>'
'Indication': 'experiment#name',
'Tissue': 'sample#tissue_type',
'Compartment': 'sc_rna_seq#biospecimen',
'Chemistry': 'sc_rna_seq#chemistry'
}
}
}
output_json(project_data, "project_data")
selected = input_json('selected_options')
magma = connect()
# Create filters for all the 'select-bys' that the value for this attribute
# must be among the options selected in the previous step.
filters = []
for target in list(selection_atts.keys()):
if len(selected[target]) > 0:
filters.append(
buildTargetPath( selection_atts[target], pdat ) + ['::in', selected[target]]
)
seq_target = parseModelAttr(pdat['seq_h5_counts_data'])
tube_records = unique(question(
magma,
[
seq_target['model'],
[ '::has', seq_target['attribute']],
*filters,
'::all', '::identifier'
]
))
if len(tube_records) < 1:
raise RuntimeError('No records with data meet the selected criteria.')
output_json(tube_records, 'tube_recs')
from archimedes.functions.dataflow import input_path, output_json, input_json
from archimedes.functions.scanpy import scanpy as sc
from archimedes.functions.environment import project_name
scdata = sc.read(input_path('leiden_anndata.h5ad'))
pdat = input_json("project_data")[project_name]
color_options = {
'Gene': dict([[gene_id, None] for gene_id in scdata.raw.var_names]),
'Cluster': None,
'Manual Annotations': None,
'Tube': None
}
color_options.update([[label, None] for label in pdat['color_options'].keys()])
output_json(color_options, 'color_options')
from archimedes.functions.dataflow import output_json, input_var, input_json, parseModelAttr, buildTargetPath
from archimedes.functions.magma import connect, question
from archimedes.functions.environment import token, magma_host, project_name
input_records = input_json('record_ids')
pdat = input_json("project_data")[project_name]
seq_target = parseModelAttr(pdat['seq_h5_counts_data'])
magma = connect()
h5_locations = question(
magma,
[
seq_target['model'],
['::identifier', '::in', input_records],
'::all', seq_target['attribute'], '::url'
],
strip_identifiers=False)
output_json(h5_locations, 'h5_locations')
from archimedes.functions.dataflow import output_path, input_path, input_var, input_bool, output_json
from archimedes.functions.scanpy import scanpy as sc
scdata = sc.read(input_path('umap_anndata.h5ad'))
use_weights = input_bool('use_weights')
# Calculate leiden clustering
leiden_resolution = float(input_var('leiden_resolution'))
sc.tl.leiden(scdata,
resolution=leiden_resolution,
key_added="leiden",
use_weights=use_weights)
### Output
scdata.write(output_path('leiden_anndata.h5ad'))
clusts = scdata.obs['leiden'].tolist()
output_json(clusts, 'leiden.json')
output_json(
dict([str(clust), str(clust)]
for clust in range(max([int(str) for str in clusts]) + 1)),
'blank_annots.json')
)
DF['cluster'] = cluster
return DF
pd.concat(
(DF_per_cluster(DEdat, cluster)) for cluster in unique(scdata.obs['leiden'])
).to_csv(output_path('diffexp.csv'))
# Extract top 10 markers per cluster, ignoring given certain prefixes
names = DEdat['names']
out_len = 10
filters = input_var('ignore_prefixes').lower()
if len(filters) > 0:
def trim_prefix(list, prefix):
for item in list[:]:
if item.lower().startswith(prefix):
list.remove(item)
return list
def trim_prefixes(list, prefixes):
for prefix in prefixes:
list = trim_prefix(list, prefix)
return list
top10 = dict( [
clust,
trim_prefixes(list(names[clust]), filters.split(","))[:(out_len-1)]
] for clust in unique(scdata.obs['leiden']) )
else:
top10 = pd.DataFrame(names).head(10).to_dict('list')
output_json(top10, 'top10.json')