25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347 | def rescale (adata,
gate=None,
log=True,
imageid='imageid',
failed_markers=None,
method='all',
verbose=True,
random_state=0):
"""
Parameters:
adata (AnnData Object, required):
An annotated data object that contains single-cell expression data.
gate (DataFrame, optional):
A pandas DataFrame where the first column lists markers, and subsequent columns contain gate values for each image in the dataset. Column names must correspond to unique `imageid` identifiers. If a single column of gate values is provided for a dataset with multiple images, the same gate will be uniformly applied to all. If no gates are provided for specific markers, the function attempts to automatically determine gates using a Gaussian Mixture Model (GMM).
log (bool, optional):
If `True`, the data in `adata.raw.X` will be log-transformed (using log1p) before gate application. This transformation is recommended when automatic gate identification through GMM is performed, as it helps in normalizing data distributions.
imageid (str, optional):
The name of the column in `adata` that contains Image IDs. This is necessary for matching manual gates specified in the `gate` DataFrame to their respective images.
failed_markers (dict, optional):
A dictionary mapping `imageid` to markers that failed quality control. This allows for the exclusion of specific markers from the analysis based on prior visual inspection or other criteria. The dictionary can use `all` as a key to specify markers that failed across all images.
method (str, optional):
Specifies the gating strategy: `all` to pool data from all images for GMM application, or `by_image` to apply GMM separately for each image. `all` may introduce batch effects, while `by_image` requires sufficient variation within each image to distinguish negative from positive populations effectively.
random_state (int, optional):
The seed used by the random number generator for GMM. Ensures reproducibility of results.
verbose (bool, optional):
If `True`, detailed progress updates and diagnostic messages will be printed during the function's execution.
Returns:
Modified AnnData Object (AnnData):
Returns the input `adata` object with updated expression data (`adata.X`) after rescaling. The gates applied, either provided manually or determined automatically, are stored within `adata.uns['gates']`.
Example:
```python
# Example with manual gates
manual_gate = pd.DataFrame({'marker': ['CD3D', 'KI67'], 'gate': [7, 8]})
adata = sm.pp.rescale(adata, gate=manual_gate, failed_markers={'all': ['CD20', 'CD21']})
# Importing gates from a CSV
manual_gate = pd.read_csv('manual_gates.csv')
adata = sm.pp.rescale(adata, gate=manual_gate, failed_markers={'all': ['CD20', 'CD21']})
# Running without manual gates to use GMM for automatic gate determination
adata = sm.pp.rescale(adata, gate=None, failed_markers={'all': ['CD20', 'CD21']})
```
"""
#log=True; imageid='imageid'; failed_markers=None; method='all'; random_state=0
# make a copy to raw data if raw is none
if adata.raw is None:
adata.raw = adata
# Mapping between markers and gates in the given dataset
dataset_markers = adata.var.index.tolist()
dataset_images = adata.obs[imageid].unique().tolist()
m= pd.DataFrame(index=dataset_markers, columns=dataset_images).reset_index()
m= pd.melt(m, id_vars=[m.columns[0]])
m.columns = ['markers', 'imageid', 'gate']
# Manipulate m with and without provided manual fates
if gate is None:
gate_mapping = m.copy()
elif bool(set(list(gate.columns)) & set(dataset_images)) is False:
global_manual_m = pd.melt(gate, id_vars=[gate.columns[0]])
global_manual_m.columns = ['markers', 'imageid', 'm_gate']
gate_mapping = m.copy()
gate_mapping.gate = gate_mapping.gate.fillna(gate_mapping.markers.map(dict(zip(global_manual_m.markers, global_manual_m.m_gate))))
else:
manual_m = pd.melt(gate, id_vars=[gate.columns[0]])
manual_m.columns = ['markers', 'imageid', 'm_gate']
gate_mapping = pd.merge(m, manual_m, how='left', left_on=['markers','imageid'], right_on = ['markers','imageid'])
gate_mapping['gate'] = gate_mapping['gate'].fillna(gate_mapping['m_gate'])
gate_mapping = gate_mapping.drop(columns='m_gate')
# Addressing failed markers
def process_failed (adata_subset, foramted_failed_markers):
if verbose:
print('Processing Failed Marker in ' + str(adata_subset.obs[imageid].unique()[0]))
# prepare data
data_subset = pd.DataFrame(adata_subset.raw.X, columns=adata_subset.var.index, index=adata_subset.obs.index)
if log is True:
data_subset = np.log1p(data_subset)
# subset markers in the subset
fm_sub = foramted_failed_markers[adata_subset.obs[imageid].unique()].dropna()
def process_failed_internal (fail_mark, data_subset):
return data_subset[fail_mark].max()
r_process_failed_internal = lambda x: process_failed_internal (fail_mark=x,data_subset=data_subset)
f_g = list(map(r_process_failed_internal, [ x[0] for x in fm_sub.values]))
subset_gate = pd.DataFrame( {'markers': [ x[0] for x in fm_sub.values],
'imageid': adata_subset.obs[imageid].unique()[0],
'gate': f_g,})
# return
return subset_gate
# Identify the failed markers
if failed_markers is not None:
# check if failed marker is a dict
if isinstance(failed_markers, dict) is False:
raise ValueError ('`failed_markers` should be a python dictionary, please refer documentation')
# create a copy
fm = failed_markers.copy()
# seperate all from the rest
if 'all' in failed_markers:
all_failed = failed_markers['all']
if isinstance(all_failed, str):
all_failed = [all_failed]
failed_markers.pop('all', None)
df = pd.DataFrame(columns = adata.obs[imageid].unique())
for i in range(len(all_failed)):
df.loc[i] = np.repeat(all_failed[i], len(df.columns))
#for i in range(len(df.columns)):
# df.loc[i] = all_failed[i]
# rest of the failed markers
#fail = pd.DataFrame.from_dict(failed_markers)
fail = pd.DataFrame(dict([ (k,pd.Series(v)) for k,v in failed_markers.items() ]))
# merge
if 'all' in fm:
foramted_failed_markers = pd.concat([fail, df], axis=0)
else:
foramted_failed_markers = fail
# send the adata objects that need to be processed
# Check if any image needs to pass through the GMM protocol
adata_list = [adata[adata.obs[imageid] == i] for i in foramted_failed_markers.columns]
# apply the process_failed function
r_process_failed = lambda x: process_failed (adata_subset=x,foramted_failed_markers=foramted_failed_markers)
failed_gates = list(map(r_process_failed, adata_list))
# combine the results and merge with gate_mapping
result = []
for i in range(len(failed_gates)):
result.append(failed_gates[i])
result = pd.concat(result, join='outer')
# use this to merge with gate_mapping
x1 = gate_mapping.set_index(['markers', 'imageid'])['gate']
x2 = result.set_index(['markers', 'imageid'])['gate']
x1.update(x2)
gate_mapping = x1.reset_index()
# trim the data before applying GMM
def clipping (x):
clip = x.clip(lower =np.percentile(x,0.01), upper=np.percentile(x,99.99)).tolist()
return clip
# Find GMM based gates
def gmm_gating (marker, data):
"""Internal function to identify gates using GMM"""
# Prepare data for GMM
data_gm = data[marker].values.reshape(-1, 1)
data_gm = data_gm[~np.isnan(data_gm)]
# Fit GMM with 3 components
gmm = GaussianMixture(n_components=3, random_state=random_state).fit(data_gm)
# Sort components by their means
means = gmm.means_.flatten()
sorted_idx = np.argsort(means)
sorted_means = means[sorted_idx]
# Calculate gate as midpoint between middle and high components
gate = np.mean([sorted_means[1], sorted_means[2]])
return gate
# Running gmm_gating on the dataset
def gmm_gating_internal (adata_subset, gate_mapping, method):
if verbose:
print('GMM for ' + str(adata_subset.obs[imageid].unique()))
data_subset = pd.DataFrame(adata_subset.raw.X, columns=adata_subset.var.index, index=adata_subset.obs.index)
# find markers
if method == 'all':
image_specific = gate_mapping.copy()
marker_to_gate = list(gate_mapping[gate_mapping.gate.isnull()].markers.unique())
else:
image_specific = gate_mapping[gate_mapping['imageid'].isin(adata_subset.obs[imageid].unique())]
marker_to_gate = image_specific[image_specific.gate.isnull()].markers.values
# Apply clipping
data_subset_clipped = data_subset.apply(clipping)
# log transform data
if log is True:
data_subset_clipped = np.log1p(data_subset_clipped)
# identify the gates for the markers
r_gmm_gating = lambda x: gmm_gating(marker=x, data=data_subset_clipped)
gates = list(map(r_gmm_gating, marker_to_gate))
# create a df with results
result = image_specific[image_specific.gate.isnull()]
mapping = dict(zip(marker_to_gate, gates))
for i in result.index:
result.loc[i, 'gate'] = mapping[result.loc[i, 'markers']]
#result['gate'] = result['gate'].fillna(result['markers'].map(dict(zip(marker_to_gate, gates))))
# return
return result
# Create a list of image IDs that need to go through the GMM
gmm_images = gate_mapping[gate_mapping.gate.isnull()].imageid.unique()
# Check if any image needs to pass through the GMM protocol
if len(gmm_images) > 0 :
# Create a list of adata that need to go through the GMM
if method == 'all':
adata_list = [adata]
else:
adata_list = [adata[adata.obs[imageid] == i] for i in gmm_images]
# run function
r_gmm_gating_internal = lambda x: gmm_gating_internal (adata_subset=x,
gate_mapping=gate_mapping,
method=method)
all_gates = list(map(r_gmm_gating_internal, adata_list))
# combine the results and merge with gate_mapping
result = []
for i in range(len(all_gates)):
result.append(all_gates[i])
result = pd.concat(result, join='outer')
# use this to merge with gate_mapping
gate_mapping.gate = gate_mapping.gate.fillna(gate_mapping.markers.map(dict(zip(result.markers, result.gate))))
# Rescaling function
def data_scaler (adata_subset, gate_mapping):
if verbose:
print('Scaling Image ' + str(adata_subset.obs[imageid].unique()[0]))
# Organise data
data_subset = pd.DataFrame(adata_subset.raw.X, columns=adata_subset.var.index, index=adata_subset.obs.index)
if log is True:
data_subset = np.log1p(data_subset)
# subset markers in the subset
gate_mapping_sub = gate_mapping[gate_mapping['imageid'] == adata_subset.obs[imageid].unique()[0]]
# organise gates
def data_scaler_internal (marker, gate_mapping_sub):
if verbose:
print('Scaling ' + str(marker))
# find the gate
moi = gate_mapping_sub[gate_mapping_sub.markers == marker]['gate'].values[0]
# Find the closest value to the gate
absolute_val_array = np.abs(data_subset[marker].values - float(moi))
# throw error if the array has nan values
if np.isnan(absolute_val_array).any():
raise ValueError ("An exception occurred: " + str(marker) + ' has nan values')
# smallest diff
smallest_difference_index = absolute_val_array.argmin()
closest_element = data_subset[marker].values[smallest_difference_index]
# rescale the data based on the identified gate
marker_study = data_subset[marker]
marker_study = marker_study.sort_values(axis=0)
# Find the index of the gate
# account for 0
if all(marker_study == 0):
gate_index = pd.DataFrame(marker_study).tail(2).index[0]
else:
gate_index = marker_study.index[marker_study == closest_element][0]
# Split into high and low groups
high = marker_study[gate_index:]
low = marker_study[:gate_index]
# Prepare for scaling the high and low dataframes
scaler_high = MinMaxScaler(feature_range=(0.5, 1))
scaler_low = MinMaxScaler(feature_range=(0, 0.5))
# Scale it
h = pd.DataFrame(scaler_high.fit_transform(high.values.reshape(-1, 1)), index = high.index)
l = pd.DataFrame(scaler_low.fit_transform(low.values.reshape(-1, 1)), index = low.index)
# Merge the high and low and resort it
scaled_data = pd.concat([l,h])
scaled_data = scaled_data.loc[~scaled_data.index.duplicated(keep='first')]
scaled_data = scaled_data.reindex(data_subset.index)
#scaled_data[scaled_data > 0.5].count(axis=1).sum()
# return
return scaled_data
# run internal function
r_data_scaler_internal = lambda x: data_scaler_internal (marker=x, gate_mapping_sub=gate_mapping_sub)
scaled_subset = list(map(r_data_scaler_internal, gate_mapping_sub.markers.values))
# combine the results and merge with gate_mapping
scaled_subset_result = []
for i in range(len(scaled_subset)):
scaled_subset_result.append(scaled_subset[i])
scaled_subset_result = pd.concat(scaled_subset_result, join='outer', axis=1)
scaled_subset_result.columns = gate_mapping_sub.markers.values
#scaled_subset_result[scaled_subset_result['CD3E'] > 0.5]['CD3E'].count(axis=1).sum()
# return
return scaled_subset_result
# pass each dataset seperately
adata_list = [adata[adata.obs[imageid] == i] for i in adata.obs[imageid].unique()]
# Run the scaler function
r_data_scaler = lambda x: data_scaler (adata_subset=x, gate_mapping=gate_mapping)
scaled_subset = list(map(r_data_scaler, adata_list))
# combine the results and merge with gate_mapping
final_result = []
for i in range(len(scaled_subset)):
final_result.append(scaled_subset[i])
final_result = pd.concat(final_result, join='outer')
# reindex the final_results
final_result = final_result.reindex(adata.obs.index)
# save final gates
adata.uns['gates'] = gate_mapping.pivot_table(index=['markers'], columns=['imageid']).droplevel(0, axis=1)#.reset_index()
# add to the anndata
adata.X = final_result
# return adata
return adata
|