davidjwen commited on
Commit
3772986
1 Parent(s): 0710c44

Added comparison to null distribution for stats

Browse files

Added comparison to null distribution for stats.
Also made some small changes to the code organization.

Files changed (1) hide show
  1. in_silico_perturber_stats.py +337 -0
in_silico_perturber_stats.py ADDED
@@ -0,0 +1,337 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Geneformer in silico perturber stats generator.
3
+
4
+ Usage:
5
+ from geneformer import InSilicoPerturberStats
6
+ ispstats = InSilicoPerturberStats(mode="goal_state_shift",
7
+ combos=0,
8
+ anchor_gene=None,
9
+ cell_states_to_model={"disease":(["dcm"],["ctrl"],["hcm"])})
10
+ ispstats.get_stats("path/to/input_data",
11
+ None,
12
+ "path/to/output_directory",
13
+ "output_prefix")
14
+ """
15
+
16
+
17
+ import os
18
+ import logging
19
+ import numpy as np
20
+ import pandas as pd
21
+ import pickle
22
+ import statsmodels.stats.multitest as smt
23
+ from pathlib import Path
24
+ from scipy.stats import ranksums
25
+ from tqdm.notebook import trange
26
+
27
+ from .tokenizer import TOKEN_DICTIONARY_FILE
28
+
29
+ GENE_NAME_ID_DICTIONARY_FILE = Path(__file__).parent / "gene_name_id_dict.pkl"
30
+
31
+ logger = logging.getLogger(__name__)
32
+
33
+ # invert dictionary keys/values
34
+ def invert_dict(dictionary):
35
+ return {v: k for k, v in dictionary.items()}
36
+
37
+ # read raw dictionary files
38
+ def read_dictionaries(dir, cell_or_gene_emb):
39
+ dict_list = []
40
+ for file in os.listdir(dir):
41
+ # process only _raw.pickle files
42
+ if file.endswith("_raw.pickle"):
43
+ with open(f"{dir}/{file}", "rb") as fp:
44
+ cos_sims_dict = pickle.load(fp)
45
+ if cell_or_gene_emb == "cell":
46
+ cell_emb_dict = {k: v for k,
47
+ v in cos_sims_dict.items() if v and "cell_emb" in k}
48
+ dict_list += [cell_emb_dict]
49
+ return dict_list
50
+
51
+ # get complete gene list
52
+ def get_gene_list(dict_list):
53
+ gene_set = set()
54
+ for dict_i in dict_list:
55
+ gene_set.update([k[0] for k, v in dict_i.items() if v])
56
+ gene_list = list(gene_set)
57
+ gene_list.sort()
58
+ return gene_list
59
+
60
+ def n_detections(token, dict_list):
61
+ cos_sim_megalist = []
62
+ for dict_i in dict_list:
63
+ cos_sim_megalist += dict_i.get((token, "cell_emb"),[])
64
+ return len(cos_sim_megalist)
65
+
66
+ def get_fdr(pvalues):
67
+ return list(smt.multipletests(pvalues, alpha=0.05, method="fdr_bh")[1])
68
+
69
+ def isp_stats(cos_sims_df, dict_list, cell_states_to_model):
70
+ random_tuples = []
71
+ for i in trange(cos_sims_df.shape[0]):
72
+ token = cos_sims_df["Gene"][i]
73
+ for dict_i in dict_list:
74
+ random_tuples += dict_i.get((token, "cell_emb"),[])
75
+ goal_end_random_megalist = [goal_end for goal_end,alt_end,start_state in random_tuples]
76
+ alt_end_random_megalist = [alt_end for goal_end,alt_end,start_state in random_tuples]
77
+ start_state_random_megalist = [start_state for goal_end,alt_end,start_state in random_tuples]
78
+
79
+ # downsample to improve speed of ranksums
80
+ if len(goal_end_random_megalist) > 100_000:
81
+ random.seed(42)
82
+ goal_end_random_megalist = random.sample(goal_end_random_megalist, k=100_000)
83
+ if len(alt_end_random_megalist) > 100_000:
84
+ random.seed(42)
85
+ alt_end_random_megalist = random.sample(alt_end_random_megalist, k=100_000)
86
+ if len(start_state_random_megalist) > 100_000:
87
+ random.seed(42)
88
+ start_state_random_megalist = random.sample(start_state_random_megalist, k=100_000)
89
+
90
+ names=["Gene",
91
+ "Gene_name",
92
+ "Ensembl_ID",
93
+ "Shift_from_goal_end",
94
+ "Shift_from_alt_end",
95
+ "Goal_end_vs_random_pval",
96
+ "Alt_end_vs_random_pval"]
97
+ cos_sims_full_df = pd.DataFrame(columns=names)
98
+
99
+ for i in trange(cos_sims_df.shape[0]):
100
+ token = cos_sims_df["Gene"][i]
101
+ name = cos_sims_df["Gene_name"][i]
102
+ ensembl_id = cos_sims_df["Ensembl_ID"][i]
103
+ token_tuples = []
104
+
105
+ for dict_i in dict_list:
106
+ token_tuples += dict_i.get((token, "cell_emb"),[])
107
+
108
+ goal_end_cos_sim_megalist = [goal_end for goal_end,alt_end,start_state in token_tuples]
109
+ alt_end_cos_sim_megalist = [alt_end for goal_end,alt_end,start_state in token_tuples]
110
+
111
+ mean_goal_end = np.mean(goal_end_cos_sim_megalist)
112
+ mean_alt_end = np.mean(alt_end_cos_sim_megalist)
113
+
114
+ pval_goal_end = ranksums(goal_end_random_megalist,goal_end_cos_sim_megalist).pvalue
115
+ pval_alt_end = ranksums(alt_end_random_megalist,alt_end_cos_sim_megalist).pvalue
116
+
117
+ data_i = [token,
118
+ name,
119
+ ensembl_id,
120
+ mean_goal_end,
121
+ mean_alt_end,
122
+ pval_goal_end,
123
+ pval_alt_end]
124
+
125
+ cos_sims_df_i = pd.DataFrame(dict(zip(names,data_i)),index=[i])
126
+ cos_sims_full_df = pd.concat([cos_sims_full_df,cos_sims_df_i])
127
+
128
+ cos_sims_full_df["Goal_end_FDR"] = get_fdr(list(cos_sims_full_df["Goal_end_vs_random_pval"]))
129
+ cos_sims_full_df["Alt_end_FDR"] = get_fdr(list(cos_sims_full_df["Alt_end_vs_random_pval"]))
130
+
131
+ return cos_sims_full_df
132
+
133
+ def isp_stats_vs_null(cos_sims_df, dict_list, null_dict_list):
134
+ cos_sims_full_df = cos_sims_df.copy()
135
+
136
+ # I think pre-initializing is faster than concatenating
137
+ cos_sims_full_df["Shift_avg"] = np.empty(cos_sims_df.shape[0], dtype=float)
138
+ cos_sims_full_df["Shift_pval"] = np.empty(cos_sims_df.shape[0], dtype=float)
139
+ cos_sims_full_df["Null_avg"] = np.empty(cos_sims_df.shape[0], dtype=float)
140
+ cos_sims_full_df["N_Detections"] = np.empty(cos_sims_df.shape[0], dtype="uint_32")
141
+ cos_sims_full_df["N_Detections_null"] = np.empty(cos_sims_df.shape[0], dtype="uint_32")
142
+
143
+ for i in trange(cos_sims_df.shape[0]):
144
+ token = cos_sims_df["Gene"][i]
145
+ name = cos_sims_df["Gene_name"][i]
146
+ ensembl_id = cos_sims_df["Ensembl_ID"][i]
147
+ token_shifts = []
148
+ null_shifts = []
149
+
150
+ for dict_i in dict_list:
151
+ token_tuples += dict_i.get((token, "cell_emb"),[])
152
+
153
+ for dict_i in null_dict_list:
154
+ null_tuples += dict_i.get((token, "cell_emb"),[])
155
+
156
+ cos_sims_full_df.loc[i, "Shift_pvalue"] = ranksums(token_shifts,
157
+ null_shifts, nan_policy="omit").pvalue
158
+ cos_sims_full_df.loc[i, "Shift_avg"] = np.mean(token_shifts)
159
+ cos_sims_full_df.loc[i, "Null_avg"] = np.mean(null_shifts)
160
+ cos_sims_full_df.loc[i, "N_Detections"] = len(token_shifts)
161
+ cos_sims_full_df.loc[i, "N_Detections_null"] = len(null_shifts)
162
+
163
+ cos_sims_full_df["Shift_FDR"] = get_fdr(cos_sims_full_df["Shift_pvalue"])
164
+ return cos_sims_full_df
165
+
166
+ class InSilicoPerturberStats:
167
+ valid_option_dict = {
168
+ "mode": {"goal_state_shift","vs_null","vs_random"},
169
+ "combos": {0,1,2},
170
+ "anchor_gene": {None, str},
171
+ "cell_states_to_model": {None, dict},
172
+ }
173
+ def __init__(
174
+ self,
175
+ mode="vs_random",
176
+ combos=0,
177
+ anchor_gene=None,
178
+ cell_states_to_model=None,
179
+ token_dictionary_file=TOKEN_DICTIONARY_FILE,
180
+ gene_name_id_dictionary_file=GENE_NAME_ID_DICTIONARY_FILE,
181
+ ):
182
+ """
183
+ Initialize in silico perturber stats generator.
184
+
185
+ Parameters
186
+ ----------
187
+ mode : {"goal_state_shift","vs_null","vs_random"}
188
+ Type of stats.
189
+ "goal_state_shift": perturbation vs. random for desired cell state shift
190
+ "vs_null": perturbation vs. null from provided null distribution dataset
191
+ "vs_random": perturbation vs. random gene perturbations in that cell (no goal direction)
192
+ combos : {0,1,2}
193
+ Whether to perturb genes individually (0), in pairs (1), or in triplets (2).
194
+ anchor_gene : None, str
195
+ ENSEMBL ID of gene to use as anchor in combination perturbations.
196
+ For example, if combos=1 and anchor_gene="ENSG00000148400":
197
+ anchor gene will be perturbed in combination with each other gene.
198
+ cell_states_to_model: None, dict
199
+ Cell states to model if testing perturbations that achieve goal state change.
200
+ Single-item dictionary with key being cell attribute (e.g. "disease").
201
+ Value is tuple of three lists indicating start state, goal end state, and alternate possible end states.
202
+ token_dictionary_file : Path
203
+ Path to pickle file containing token dictionary (Ensembl ID:token).
204
+ gene_name_id_dictionary_file : Path
205
+ Path to pickle file containing gene name to ID dictionary (gene name:Ensembl ID).
206
+ """
207
+
208
+ self.mode = mode
209
+ self.combos = combos
210
+ self.anchor_gene = anchor_gene
211
+ self.cell_states_to_model = cell_states_to_model
212
+
213
+ self.validate_options()
214
+
215
+ # load token dictionary (Ensembl IDs:token)
216
+ with open(token_dictionary_file, "rb") as f:
217
+ self.gene_token_dict = pickle.load(f)
218
+
219
+ # load gene name dictionary (gene name:Ensembl ID)
220
+ with open(gene_name_id_dictionary_file, "rb") as f:
221
+ self.gene_name_id_dict = pickle.load(f)
222
+
223
+ if anchor_gene is None:
224
+ self.anchor_token = None
225
+ else:
226
+ self.anchor_token = self.gene_token_dict[self.anchor_gene]
227
+
228
+ def validate_options(self):
229
+ for attr_name,valid_options in self.valid_option_dict.items():
230
+ attr_value = self.__dict__[attr_name]
231
+ if type(attr_value) not in {list, dict}:
232
+ if attr_value in valid_options:
233
+ continue
234
+ valid_type = False
235
+ for option in valid_options:
236
+ if (option in [int,list,dict]) and isinstance(attr_value, option):
237
+ valid_type = True
238
+ break
239
+ if valid_type:
240
+ continue
241
+ logger.error(
242
+ f"Invalid option for {attr_name}. " \
243
+ f"Valid options for {attr_name}: {valid_options}"
244
+ )
245
+ raise
246
+
247
+ if self.cell_states_to_model is not None:
248
+ if (len(self.cell_states_to_model.items()) == 1):
249
+ for key,value in self.cell_states_to_model.items():
250
+ if (len(value) == 3) and isinstance(value, tuple):
251
+ if isinstance(value[0],list) and isinstance(value[1],list) and isinstance(value[2],list):
252
+ if len(value[0]) == 1 and len(value[1]) == 1:
253
+ all_values = value[0]+value[1]+value[2]
254
+ if len(all_values) == len(set(all_values)):
255
+ continue
256
+ else:
257
+ logger.error(
258
+ "Cell states to model must be a single-item dictionary with " \
259
+ "key being cell attribute (e.g. 'disease') and value being " \
260
+ "tuple of three lists indicating start state, goal end state, and alternate possible end states. " \
261
+ "Values should all be unique. " \
262
+ "For example: {'disease':(['start_state'],['ctrl'],['alt_end'])}")
263
+ raise
264
+ if self.anchor_gene is not None:
265
+ self.anchor_gene = None
266
+ logger.warning(
267
+ "anchor_gene set to None. " \
268
+ "Currently, anchor gene not available " \
269
+ "when modeling multiple cell states.")
270
+
271
+ def get_stats(self,
272
+ input_data_directory,
273
+ null_dist_data_directory,
274
+ output_directory,
275
+ output_prefix):
276
+ """
277
+ Get stats for in silico perturbation data and save as results in output_directory.
278
+
279
+ Parameters
280
+ ----------
281
+ input_data_directory : Path
282
+ Path to directory containing cos_sim dictionary inputs
283
+ null_dist_data_directory : Path
284
+ Path to directory containing null distribution cos_sim dictionary inputs
285
+ output_directory : Path
286
+ Path to directory where perturbation data will be saved as .csv
287
+ output_prefix : str
288
+ Prefix for output .dataset
289
+ """
290
+
291
+ if self.mode not in ["goal_state_shift", "vs_null"]:
292
+ logger.error(
293
+ "Currently, only modes available are stats for goal_state_shift \
294
+ and comparing vs a null distribution.")
295
+ raise
296
+
297
+ self.gene_token_id_dict = invert_dict(self.gene_token_dict)
298
+ self.gene_id_name_dict = invert_dict(self.gene_name_id_dict)
299
+
300
+ # obtain total gene list
301
+ gene_list = get_gene_list(dict_list)
302
+
303
+ # initiate results dataframe
304
+ cos_sims_df_initial = pd.DataFrame({"Gene": gene_list,
305
+ "Gene_name": [self.token_to_gene_name(item) \
306
+ for item in gene_list], \
307
+ "Ensembl_ID": [self.gene_token_id_dict[genes[1]] \
308
+ if isinstance(genes,tuple) else \
309
+ self.gene_token_id_dict[genes] \
310
+ for genes in gene_list]}, \
311
+ index=[i for i in range(len(gene_list))])
312
+
313
+ dict_list = read_dictionaries(input_data_directory, "cell")
314
+ if self.mode == "goal_state_shift":
315
+ cos_sims_df = isp_stats(cos_sims_df_initial, dict_list, self.cell_states_to_model)
316
+
317
+ # quantify number of detections of each gene
318
+ cos_sims_df["N_Detections"] = [n_detections(i, dict_list) for i in cos_sims_df["Gene"]]
319
+
320
+ # sort by shift to desired state
321
+ cos_sims_df = cos_sims_df.sort_values(by=["Shift_from_goal_end",
322
+ "Goal_end_FDR"])
323
+ elif self.mode == "vs_null":
324
+ dict_list = read_dictionaries(input_data_directory, "cell")
325
+ null_dict_list = read_dictionaries(null_dist_data_directory, "cell")
326
+ cos_sims_df = isp_stats_vs_null(cos_sims_df_initial, dict_list,
327
+ null_dict_list)
328
+
329
+ # save perturbation stats to output_path
330
+ output_path = (Path(output_directory) / output_prefix).with_suffix(".csv")
331
+ cos_sims_df.to_csv(output_path)
332
+
333
+ def token_to_gene_name(self, item):
334
+ if isinstance(item,int):
335
+ return self.gene_id_name_dict.get(self.gene_token_id_dict.get(item, np.nan), np.nan)
336
+ if isinstance(item,tuple):
337
+ return tuple([self.gene_id_name_dict.get(self.gene_token_id_dict.get(i, np.nan), np.nan) for i in item])