-
Notifications
You must be signed in to change notification settings - Fork 180
/
transform.py
471 lines (407 loc) · 19.4 KB
/
transform.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
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
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
# -*- coding: utf-8 -*-
from ctxcore.recovery import recovery, aucs as calc_aucs
import logging
import traceback
import pandas as pd
import numpy as np
from .utils import (
COLUMN_NAME_MOTIF_SIMILARITY_QVALUE,
COLUMN_NAME_ORTHOLOGOUS_IDENTITY,
COLUMN_NAME_MOTIF_ID,
COLUMN_NAME_TF,
COLUMN_NAME_ANNOTATION,
ACTIVATING_MODULE,
REPRESSING_MODULE,
)
from itertools import repeat
from ctxcore.rnkdb import RankingDatabase
from functools import reduce
from typing import Type, Sequence, Optional
from ctxcore.genesig import Regulon
from ctxcore.recovery import leading_edge4row
import math
from itertools import chain
from functools import partial
from cytoolz import first
import numpy as np
from dask.dataframe.utils import make_meta
COLUMN_NAME_NES = "NES"
COLUMN_NAME_AUC = "AUC"
COLUMN_NAME_CONTEXT = "Context"
COLUMN_NAME_TARGET_GENES = "TargetGenes"
COLUMN_NAME_RANK_AT_MAX = "RankAtMax"
COLUMN_NAME_TYPE = "Type"
# TODO: Should actually be a function depending on return_recovery_curves and rank_threshold
DF_META_DATA = make_meta(
{
('Enrichment', COLUMN_NAME_AUC): np.float64,
('Enrichment', COLUMN_NAME_NES): np.float64,
('Enrichment', COLUMN_NAME_MOTIF_SIMILARITY_QVALUE): np.float64,
('Enrichment', COLUMN_NAME_ORTHOLOGOUS_IDENTITY): np.float64,
('Enrichment', COLUMN_NAME_ANNOTATION): np.object,
('Enrichment', COLUMN_NAME_CONTEXT): np.object,
('Enrichment', COLUMN_NAME_TARGET_GENES): np.object,
('Enrichment', COLUMN_NAME_RANK_AT_MAX): np.int64,
},
index=pd.MultiIndex.from_arrays([[], []], names=(COLUMN_NAME_TF, COLUMN_NAME_MOTIF_ID)),
)
__all__ = ["module2features", "module2df", "modules2df", "df2regulons", "module2regulon", "modules2regulons"]
LOGGER = logging.getLogger(__name__)
def module2features_rcc4all_impl(
db: Type[RankingDatabase],
module: Regulon,
motif_annotations: pd.DataFrame,
rank_threshold: int = 1500,
auc_threshold: float = 0.05,
nes_threshold=3.0,
weighted_recovery=False,
filter_for_annotation=True,
):
"""
Create a dataframe of enriched and annotated features a given ranking database and a co-expression module.
:param db: The ranking database.
:param module: The co-expression module.
:param rank_threshold: The total number of ranked genes to take into account when creating a recovery curve.
:param auc_threshold: The fraction of the ranked genome to take into account for the calculation of the
Area Under the recovery Curve.
:param nes_threshold: The Normalized Enrichment Score (NES) threshold to select enriched features.
:param weighted_recovery: Use weighted recovery in the analysis.
:return: A dataframe with enriched and annotated features.
"""
# Load rank of genes from database.
df = db.load(module)
features, genes, rankings = df.index.values, df.columns.values, df.values
weights = np.asarray([module[gene] for gene in genes]) if weighted_recovery else np.ones(len(genes))
# Calculate recovery curves, AUC and NES values.
rccs, aucs = recovery(df, db.total_genes, weights, rank_threshold, auc_threshold)
ness = (aucs - aucs.mean()) / aucs.std()
# Keep only features that are enriched, i.e. NES sufficiently high.
enriched_features_idx = ness >= nes_threshold
enriched_features = pd.DataFrame(
index=pd.MultiIndex.from_tuples(
list(zip(repeat(module.transcription_factor), features[enriched_features_idx])),
names=[COLUMN_NAME_TF, COLUMN_NAME_MOTIF_ID],
),
data={COLUMN_NAME_NES: ness[enriched_features_idx], COLUMN_NAME_AUC: aucs[enriched_features_idx]},
)
if len(enriched_features) == 0:
return pd.DataFrame(), None, None, genes, None
# Find motif annotations for enriched features.
annotated_features = pd.merge(enriched_features, motif_annotations, how="left", left_index=True, right_index=True)
annotated_features_idx = (
pd.notnull(annotated_features[COLUMN_NAME_ANNOTATION])
if filter_for_annotation
else np.full((len(enriched_features),), True)
)
if len(annotated_features[annotated_features_idx]) == 0:
return pd.DataFrame(), None, None, genes, None
# Calculated leading edge for the remaining enriched features that have annotations.
avgrcc = rccs.mean(axis=0)
avg2stdrcc = avgrcc + 2.0 * rccs.std(axis=0)
rccs = rccs[enriched_features_idx, :][annotated_features_idx, :]
rankings = rankings[enriched_features_idx, :][annotated_features_idx, :]
# Add additional information to the dataframe.
annotated_features = annotated_features[annotated_features_idx]
context = frozenset(chain(module.context, [db.name]))
annotated_features[COLUMN_NAME_CONTEXT] = len(annotated_features) * [context]
return annotated_features, rccs, rankings, genes, avg2stdrcc
def module2features_auc1st_impl(
db: Type[RankingDatabase],
module: Regulon,
motif_annotations: pd.DataFrame,
rank_threshold: int = 1500,
auc_threshold: float = 0.05,
nes_threshold=3.0,
weighted_recovery=False,
filter_for_annotation=True,
):
"""
Create a dataframe of enriched and annotated features a given ranking database and a co-expression module.
:param db: The ranking database.
:param module: The co-expression module.
:param rank_threshold: The total number of ranked genes to take into account when creating a recovery curve.
:param auc_threshold: The fraction of the ranked genome to take into account for the calculation of the
Area Under the recovery Curve.
:param nes_threshold: The Normalized Enrichment Score (NES) threshold to select enriched features.
:param weighted_recovery: Use weighted recovery in the analysis.
:return: A dataframe with enriched and annotated features.
"""
# Load rank of genes from database.
df = db.load(module)
features, genes, rankings = df.index.values, df.columns.values, df.values
weights = np.asarray([module[gene] for gene in genes]) if weighted_recovery else np.ones(len(genes))
# include check for modules with no genes that could be mapped to the db. This can happen when including non protein-coding genes in the expression matrix.
if df.empty:
LOGGER.warning(
"No genes in module {} could be mapped to {}. Skipping this module.".format(module.name, db.name)
)
return pd.DataFrame(), None, None, genes, None
# Calculate recovery curves, AUC and NES values.
# For fast unweighted implementation so weights to None.
aucs = calc_aucs(df, db.total_genes, weights, auc_threshold)
ness = (aucs - aucs.mean()) / aucs.std()
# Keep only features that are enriched, i.e. NES sufficiently high.
enriched_features_idx = ness >= nes_threshold
enriched_features = pd.DataFrame(
index=pd.MultiIndex.from_tuples(
list(zip(repeat(module.transcription_factor), features[enriched_features_idx])),
names=[COLUMN_NAME_TF, COLUMN_NAME_MOTIF_ID],
),
data={COLUMN_NAME_NES: ness[enriched_features_idx], COLUMN_NAME_AUC: aucs[enriched_features_idx]},
)
if len(enriched_features) == 0:
return pd.DataFrame(), None, None, genes, None
# Find motif annotations for enriched features.
annotated_features = pd.merge(enriched_features, motif_annotations, how="left", left_index=True, right_index=True)
annotated_features_idx = (
pd.notnull(annotated_features[COLUMN_NAME_ANNOTATION])
if filter_for_annotation
else np.full((len(enriched_features),), True)
)
if len(annotated_features[annotated_features_idx]) == 0:
return pd.DataFrame(), None, None, genes, None
# Calculated leading edge for the remaining enriched features that have annotations. The leading edge is calculated
# based on the average recovery curve so we still no to calculate all recovery curves. Currently this is done via
# preallocating memory. This introduces a huge burden on memory when using region-based databases and multiple cores
# on a cluster node. E.g.
# (24,000 features * 25,000 rank_threshold * 8 bytes)/(1,024*1,024*1,024) = 4,4Gb
# This creates a potential peak on memory of 48 cores * 4,4Gb = 214 Gb
# TODO: Solution could be to go for an iterative approach boosted by numba. But before doing so investigate the
# broader issue with creep in memory usage when using the dask framework: use a memory profile tool
# (https://pythonhosted.org/Pympler/muppy.html) to check what is kept in memory in all subprocesses/workers.
rccs, _ = recovery(df, db.total_genes, weights, rank_threshold, auc_threshold, no_auc=True)
avgrcc = rccs.mean(axis=0)
avg2stdrcc = avgrcc + 2.0 * rccs.std(axis=0)
rccs = rccs[enriched_features_idx, :][annotated_features_idx, :]
rankings = rankings[enriched_features_idx, :][annotated_features_idx, :]
# Add additional information to the dataframe.
annotated_features = annotated_features[annotated_features_idx]
context = frozenset(chain(module.context, [db.name]))
annotated_features[COLUMN_NAME_CONTEXT] = len(annotated_features) * [context]
return annotated_features, rccs, rankings, genes, avg2stdrcc
module2features = partial(
module2features_auc1st_impl, rank_threshold=1500, auc_threshold=0.05, nes_threshold=3.0, filter_for_annotation=True
)
def module2df(
db: Type[RankingDatabase],
module: Regulon,
motif_annotations: pd.DataFrame,
weighted_recovery=False,
return_recovery_curves=False,
module2features_func=module2features,
) -> pd.DataFrame:
""" """
# Derive enriched and TF-annotated features for module.
try:
df_annotated_features, rccs, rankings, genes, avg2stdrcc = module2features_func(
db, module, motif_annotations, weighted_recovery=weighted_recovery
)
except MemoryError:
LOGGER.error(
"Unable to process \"{}\" on database \"{}\" because ran out of memory. Stacktrace:".format(
module.name, db.name
)
)
LOGGER.error(traceback.format_exc())
return DF_META_DATA
# If less than 80% of the genes are mapped to the ranking database, the module is skipped.
n_missing = len(module) - len(genes)
frac_missing = float(n_missing) / len(module)
if frac_missing >= 0.20:
LOGGER.warning(
"Less than 80% of the genes in {} could be mapped to {}. Skipping this module.".format(module.name, db.name)
)
return DF_META_DATA
# If no annotated enriched features could be found, skip module.
if len(df_annotated_features) == 0:
return DF_META_DATA
rank_threshold = rccs.shape[1]
# Combine elements into a dataframe.
df_annotated_features.columns = pd.MultiIndex.from_tuples(
list(zip(repeat("Enrichment"), df_annotated_features.columns))
)
df_rnks = pd.DataFrame(
index=df_annotated_features.index,
columns=pd.MultiIndex.from_tuples(list(zip(repeat("Ranking"), genes))),
data=rankings,
)
df_rccs = pd.DataFrame(
index=df_annotated_features.index,
columns=pd.MultiIndex.from_tuples(list(zip(repeat("Recovery"), np.arange(rank_threshold)))),
data=rccs,
)
df = pd.concat([df_annotated_features, df_rccs, df_rnks], axis=1)
# Calculate the leading edges for each row. Always return importance from gene inference phase.
weights = np.array([module[gene] for gene in genes])
df[[("Enrichment", COLUMN_NAME_TARGET_GENES), ("Enrichment", COLUMN_NAME_RANK_AT_MAX)]] = df.apply(
partial(leading_edge4row, avg2stdrcc=avg2stdrcc, genes=genes, weights=weights), axis=1
)
# Remove unnecessary data from dataframe.
del df['Ranking']
if not return_recovery_curves:
del df['Recovery']
assert all(
[col in df.columns for col in DF_META_DATA]
), f"Column comparison to expected metadata failed! Found:\n{df.columns}"
return df[DF_META_DATA.columns]
else:
return df
def modules2df(
db: Type[RankingDatabase],
modules: Sequence[Regulon],
motif_annotations: pd.DataFrame,
weighted_recovery=False,
return_recovery_curves=False,
module2features_func=module2features,
) -> pd.DataFrame:
# Make sure return recovery curves is always set to false because the metadata for the distributed dataframe needs
# to be fixed for the dask framework.
# TODO: Remove this restriction.
return pd.concat(
[module2df(db, module, motif_annotations, weighted_recovery, False, module2features_func) for module in modules]
)
def _regulon4group(tf_name, context, df_group, save_columns=[]) -> Optional[Regulon]:
def score(nes, motif_similarity_qval, orthologuous_identity):
# The combined score starts from the NES score which is then corrected for less confidence in the TF annotation
# in two steps:
# 1. The orthologous identifity (a fraction between 0 and 1.0) is used directly to normalize the NES.
# 2. The motif similarity q-value is converted to a similar fraction: -log10(q-value)
# A motif that is directly annotated for the TF in the correct species is not penalized.
correction_fraction = 1.0
try:
max_value = 10 # A q-value smaller than 10**-10 is considered the same as a q-value of 0.0.
correction_fraction = (
min(-math.log(motif_similarity_qval, 10), max_value) / max_value
if not math.isnan(motif_similarity_qval)
else 1.0
)
except ValueError: # Math domain error
pass
score = nes * correction_fraction
# We assume that a non existing orthologous identity signifies a direct annotation.
return score if math.isnan(orthologuous_identity) else score * orthologuous_identity
def derive_interaction_type(ctx):
return "(-)" if REPRESSING_MODULE in ctx else "(+)"
def row2regulon(row):
# The target genes as well as their weights/importances are directly taken from the dataframe.
return Regulon(
name="{}{}".format(tf_name, derive_interaction_type(context)),
score=score(
row[COLUMN_NAME_NES], row[COLUMN_NAME_MOTIF_SIMILARITY_QVALUE], row[COLUMN_NAME_ORTHOLOGOUS_IDENTITY]
),
context=context,
transcription_factor=tf_name,
gene2weight=row[COLUMN_NAME_TARGET_GENES],
gene2occurrence=[],
)
# Find most enriched annotated motif and add this to the context
df_selected = df_group.sort_values(by=COLUMN_NAME_NES, ascending=False)
first_result_by_nes = df_selected.head(1).reset_index()
motif_logo = '{}.png'.format(first_result_by_nes[COLUMN_NAME_MOTIF_ID].values[0]) if len(df_selected) > 0 else ""
# Add additional columns to the regulon
nes = first_result_by_nes[COLUMN_NAME_NES].values[0] if COLUMN_NAME_NES in save_columns else 0.0
orthologous_identity = (
first_result_by_nes[COLUMN_NAME_ORTHOLOGOUS_IDENTITY].values[0]
if COLUMN_NAME_ORTHOLOGOUS_IDENTITY in save_columns
else 0.0
)
similarity_qvalue = (
first_result_by_nes[COLUMN_NAME_MOTIF_SIMILARITY_QVALUE].values[0]
if COLUMN_NAME_MOTIF_SIMILARITY_QVALUE in save_columns
else 0.0
)
annotation = first_result_by_nes[COLUMN_NAME_ANNOTATION].values[0] if COLUMN_NAME_ANNOTATION in save_columns else ''
# First we create a regulon for each enriched and annotated feature and then we aggregate these regulons into a
# single one using the union operator. This operator combined all target genes into a single set of genes keeping
# the maximum weight associated with a gene. In addition, the maximum combined score is kept as the score of the
# entire regulon.
return reduce(Regulon.union, (row2regulon(row) for _, row in df_group.iterrows())).copy(
context=frozenset(set(context).union({motif_logo})),
nes=nes,
orthologous_identity=orthologous_identity,
similarity_qvalue=similarity_qvalue,
annotation=annotation,
)
nes = first_result_by_nes[COLUMN_NAME_NES].values[0] if COLUMN_NAME_NES in save_columns else 0.0
orthologous_identity = (
first_result_by_nes[COLUMN_NAME_ORTHOLOGOUS_IDENTITY].values[0]
if COLUMN_NAME_ORTHOLOGOUS_IDENTITY in save_columns
else 0.0
)
similarity_qvalue = (
first_result_by_nes[COLUMN_NAME_MOTIF_SIMILARITY_QVALUE].values[0]
if COLUMN_NAME_MOTIF_SIMILARITY_QVALUE in save_columns
else 0.0
)
annotation = first_result_by_nes[COLUMN_NAME_ANNOTATION].values[0] if COLUMN_NAME_ANNOTATION in save_columns else ''
def df2regulons(df, save_columns=[]) -> Sequence[Regulon]:
"""
Create regulons from a dataframe of enriched features.
:param df: The dataframe.
:param save_columns: Additional columns to save from the given dataframe of enriched features. Possible values are COLUMN_NAME_NES, COLUMN_NAME_ORTHOLOGOUS_IDENTITY, COLUMN_NAME_MOTIF_SIMILARITY_QVALUE, COLUMN_NAME_ANNOTATION.
:return: A sequence of regulons.
"""
assert not df.empty, 'Signatures dataframe is empty!'
print("Create regulons from a dataframe of enriched features.")
print("Additional columns saved: {}".format(save_columns))
# Because the code below will alter the dataframe we need to make a defensive copy of it.
df = df.copy()
# Normally the columns index has two levels. For convenience of the following code the first level is removed.
if df.columns.nlevels == 2:
df.columns = df.columns.droplevel(0)
# Unpack the type of the module from the context column (dtype = frozenset)
def get_type(row):
ctx = row[COLUMN_NAME_CONTEXT]
# Activating is the default!
return REPRESSING_MODULE if REPRESSING_MODULE in ctx else ACTIVATING_MODULE
df[COLUMN_NAME_TYPE] = df.apply(get_type, axis=1)
# Group all rows per TF and type (+)/(-). Each group results in a single regulon.
not_none = lambda r: r is not None
return list(
filter(
not_none,
(
_regulon4group(tf_name, frozenset([interaction_type]), df_grp, save_columns)
for (tf_name, interaction_type), df_grp in df.groupby(by=[COLUMN_NAME_TF, COLUMN_NAME_TYPE])
),
)
)
def module2regulon(
db: Type[RankingDatabase],
module: Regulon,
motif_annotations: pd.DataFrame,
weighted_recovery=False,
return_recovery_curves=False,
module2features_func=module2features,
) -> Optional[Regulon]:
# First calculating a dataframe and then derive the regulons from them introduces a performance penalty.
df = module2df(
db,
module,
motif_annotations,
weighted_recovery=weighted_recovery,
return_recovery_curves=return_recovery_curves,
module2features_func=module2features_func,
)
if len(df) == 0:
return None
regulons = df2regulons(df)
return first(regulons) if len(regulons) > 0 else None
def modules2regulons(
db: Type[RankingDatabase],
modules: Sequence[Regulon],
motif_annotations: pd.DataFrame,
weighted_recovery=False,
return_recovery_curves=False,
module2features_func=module2features,
) -> Sequence[Regulon]:
assert len(modules) > 0
df = modules2df(
db,
modules,
motif_annotations,
weighted_recovery=weighted_recovery,
return_recovery_curves=return_recovery_curves,
module2features_func=module2features_func,
)
return [] if len(df) == 0 else df2regulons(df)