-
Notifications
You must be signed in to change notification settings - Fork 0
/
pseudocell_tracer.py
455 lines (335 loc) · 16.2 KB
/
pseudocell_tracer.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
import warnings
warnings.filterwarnings("ignore")
import os
import math
import numpy as np
import tensorflow as tf
import pandas as pd
import argparse
import json
from sklearn.model_selection import StratifiedKFold
from sklearn.preprocessing import StandardScaler
from sklearn.manifold import TSNE
from scipy.stats import spearmanr, pearsonr
import matplotlib
matplotlib.use('TkAgg')
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches
plt.rcParams["figure.figsize"] = (8,8)
import seaborn as sns
from umap import UMAP
import configparser as cp
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
###################################################
# Read in configuration file for network parameters
###################################################
config = cp.ConfigParser()
config.read('config.py')
rna_hidden_dim_1 = int(config.get('AE', 'RNA_Layer1'))
rna_hidden_dim_2 = int(config.get('AE', 'RNA_Layer2'))
rna_latent_dim = int(config.get('AE', 'RNA_Latent'))
translate_hidden_dim = int(config.get('AE', 'Translate_Layer1'))
l2_norm_dec = float(config.get('AE', 'L2_Norm_AE'))
l2_norm_kl = float(config.get('AE', 'L2_Norm_KL'))
learning_rate_kl = float(config.get('AE', 'Learning_Rate_KL'))
learning_rate_ae = float(config.get('AE', 'Learning_Rate_AE'))
Z_dim = int(config.get('CGAN', 'Z_dim'))
gen_dim1 = int(config.get('CGAN', 'Gen_Layer1'))
gen_dim2 = int(config.get('CGAN', 'Gen_Layer2'))
det_dim = int(config.get('CGAN', 'Det_Layer1'))
learning_rate_cgan = float(config.get('CGAN', 'Learning_Rate'))
l2_norm_cgan = float(config.get('CGAN', 'L2_Lambda'))
epoch_cgan = int(config.get('CGAN', 'Max_Epoch'))
###################################################
# Read in command line arguments
###################################################
parser = argparse.ArgumentParser(description='Perform Pseudocell Tracer Algorithm')
parser.add_argument('-d', '--data', help='Tab delimited file representing matrix of samples by genes', required=True)
parser.add_argument('-s', '--side_data', help= 'Tab delimited file for side information to be used', required=True)
parser.add_argument('-o', '--output', help='Output directory', required=True)
parser.add_argument('-p', '--plot_style', help= 'Plotting style to be used (UMAP or tSNE)', default="UMAP")
parser.add_argument('-n', '--num_cells_gen', help='Number of pseudocells to generate at each step', default=100, type=int)
parser.add_argument('-k', '--num_steps', help='Number of pseudocell states', default=100, type=int)
parser.add_argument('-a', '--start_states', help='Number of pseudocell states', nargs="+", required=True)
parser.add_argument('-b', '--end_states', help='Number of pseudocell states', nargs="+", required=True)
parser.add_argument('-g', '--genes_to_plot', help='Genes to plot in pseudocell trajectory', nargs="+")
args = parser.parse_args()
dset = args.data
sset = args.side_data
out = args.output
plot_method = args.plot_style
num_cells_gen = args.num_cells_gen
num_steps = args.num_steps
start_states = args.start_states
end_states = args.end_states
genes_to_plot = args.genes_to_plot
###################################################
# Function to plot
###################################################
def plot_data(rna, side):
if plot_method == "tSNE":
tsne = TSNE()
results = np.array(tsne.fit_transform(rna.values))
if plot_method == "UMAP":
umap = UMAP()
results = np.array(umap.fit_transform(rna.values))
plot_df = pd.DataFrame(data=results, columns=["X","Y"])
cl = np.argmax(side.values, axis=1)
plot_df["subtype"] = [side.columns[x] for x in cl]
return sns.scatterplot(x="X", y="Y", hue="subtype", data=plot_df)
###################################################
# Load Data
###################################################
rna_data = pd.read_csv(dset, header=0, index_col=0, sep="\t")
side_data = pd.read_csv(sset, index_col=0, header=0, sep="\t")
###################################################
# Filter common samples
###################################################
sample_list = np.intersect1d(rna_data.columns.values, side_data.columns.values)
gene_list = rna_data.index.values
side_data = side_data.filter(sample_list, axis=1)
rna_data = rna_data.filter(sample_list, axis=1)
rna_data = rna_data.transpose()
side_data = side_data.transpose()
print("Loaded Data...")
###################################################
# Create output directory
###################################################
print("Creating output directory...")
try:
os.mkdir("results")
except:
pass
out_dir = "results/" + out
try:
os.mkdir(out_dir)
print("Directory " , out_dir , " Created...")
except FileExistsError:
print("Warning: Directory " , out , " already exists...")
with open(out_dir + '/run_parameters.txt', 'w') as f:
json.dump(args.__dict__, f, indent=2)
with open(out_dir + '/run_network_config.txt', 'w') as f:
config.write(f)
###################################################
# Plot input data
###################################################
scatter = plot_data(rna_data, side_data)
plt.title("Input Data (" + str(plot_method) + ")")
plt.savefig(out_dir + "/input_scatter.png")
plt.clf()
###################################################
# Train supervised encoder
###################################################
print("Training supervised encoder...")
scaler = StandardScaler().fit(rna_data.values)
norm_rna_data = np.clip(scaler.transform(rna_data.values),-3,3)
reg_kl = tf.keras.regularizers.l2(l2_norm_kl)
kl_model = tf.keras.Sequential([
tf.keras.layers.Dense(rna_hidden_dim_1, activation='relu',
kernel_regularizer=reg_kl, input_shape=(len(gene_list),)),
tf.keras.layers.Dropout(0.3),
tf.keras.layers.Dense(rna_hidden_dim_2, activation='relu',
kernel_regularizer=reg_kl),
tf.keras.layers.Dropout(0.3),
tf.keras.layers.Dense(rna_latent_dim, activation='sigmoid', name='latent_layer',
kernel_regularizer=reg_kl),
tf.keras.layers.Dropout(0.3),
tf.keras.layers.Dense(translate_hidden_dim, activation='relu',
kernel_regularizer=reg_kl),
tf.keras.layers.Dropout(0.3),
tf.keras.layers.Dense(side_data.values.shape[1], activation=tf.nn.softmax,
kernel_regularizer=reg_kl, name='relative_prediction')
])
es_cb_kl = tf.keras.callbacks.EarlyStopping('val_kullback_leibler_divergence', patience=100, restore_best_weights=True)
kl_model.compile(optimizer=tf.train.AdamOptimizer(learning_rate_kl), loss=tf.keras.losses.KLD, metrics=['kullback_leibler_divergence'])
kl_model.fit(norm_rna_data, side_data.values, epochs=10000, callbacks=[es_cb_kl], validation_split=0.1, verbose=0, batch_size=1024)
kl_model.fit(norm_rna_data, side_data.values, epochs=5, verbose=0, batch_size=1024)
kl_model.save(out_dir + "/encoder.h5")
###################################################
# Get latent data
###################################################
print("Getting latent data...")
latent_model = tf.keras.Model(inputs=kl_model.input, outputs=kl_model.get_layer('latent_layer').output)
latent = latent_model.predict(norm_rna_data)
latent_df = pd.DataFrame(data=latent, index=sample_list)
latent_df.to_csv("latent_values.tsv", sep="\t")
scatter = plot_data(latent_df, side_data)
plt.title("Latent Data (" + str(plot_method) + ")")
plt.savefig(out_dir + "/latent_scatter.png")
plt.clf()
###################################################
# Train decoder
###################################################
print("Training decoder...")
reg_dec = tf.keras.regularizers.l2(l2_norm_dec)
dec_model = tf.keras.Sequential([
tf.keras.layers.Dense(rna_hidden_dim_2, activation='relu', input_shape=(rna_latent_dim,),
kernel_regularizer=reg_dec, bias_regularizer=reg_dec),
tf.keras.layers.BatchNormalization(),
tf.keras.layers.Dense(rna_hidden_dim_1, activation='relu',
kernel_regularizer=reg_dec, bias_regularizer=reg_dec),
tf.keras.layers.BatchNormalization(),
tf.keras.layers.Dense(len(gene_list), activation=None, name='rna_reconstruction',
kernel_regularizer=reg_dec, bias_regularizer=reg_dec)
])
es_cb_dec = tf.keras.callbacks.EarlyStopping('val_mean_squared_error', patience=100, restore_best_weights=True)
dec_model.compile(optimizer=tf.train.AdamOptimizer(learning_rate_ae), loss=tf.keras.losses.MSE, metrics=['mse'])
dec_model.fit(latent, norm_rna_data, epochs=100000, callbacks=[es_cb_dec], validation_split=0.1,
verbose=0, batch_size=1024)
dec_model.fit(latent, norm_rna_data, epochs=5, verbose=0, batch_size=1024)
dec_model.save(out_dir + "/decoder.h5")
###################################################
# Get reconstructed values
###################################################
decoded = dec_model.predict(latent)
decoded_df = pd.DataFrame(data=decoded, index=sample_list, columns=gene_list)
decoded_df.to_csv("reconstructed_values.tsv", sep="\t")
scatter = plot_data(decoded_df, side_data)
plt.title("Reconstructed Data (" + str(plot_method) + ")")
plt.savefig(out_dir + "/reconstructed_scatter.png")
plt.clf()
###################################################
# Build CGAN Architecture
###################################################
def xavier_init(size):
in_dim = size[0]
xavier_stddev = 1. / tf.sqrt(in_dim / 2.)
return tf.random_normal(shape=size, stddev=xavier_stddev)
""" Discriminator Net model """
X = tf.placeholder(tf.float32, shape=[None, rna_latent_dim])
y = tf.placeholder(tf.float32, shape=[None, side_data.shape[1]])
D_W1 = tf.Variable(xavier_init([rna_latent_dim + side_data.shape[1], det_dim]))
D_b1 = tf.Variable(tf.zeros(shape=[det_dim]))
D_W2 = tf.Variable(xavier_init([det_dim, 1]))
D_b2 = tf.Variable(tf.zeros(shape=[1]))
theta_D = [D_W1, D_W2, D_b1, D_b2]
def discriminator(x, y):
inputs = tf.concat(axis=1, values=[x, y])
D_h1 = tf.nn.relu(tf.matmul(inputs, D_W1) + D_b1)
D_logit = tf.matmul(D_h1, D_W2) + D_b2
D_prob = tf.nn.sigmoid(D_logit)
return D_prob, D_logit
""" Generator Net model """
Z = tf.placeholder(tf.float32, shape=[None, Z_dim])
G_W1 = tf.Variable(xavier_init([Z_dim + side_data.shape[1], gen_dim1]))
G_b1 = tf.Variable(tf.zeros(shape=[gen_dim1]))
G_W2 = tf.Variable(xavier_init([gen_dim1, gen_dim2]))
G_b2 = tf.Variable(tf.zeros(shape=[gen_dim2]))
G_W3 = tf.Variable(xavier_init([gen_dim2, rna_latent_dim]))
G_b3 = tf.Variable(tf.zeros(shape=[rna_latent_dim]))
theta_G = [G_W1, G_W2, G_W3, G_b1, G_b2, G_b3]
def generator(z, y):
inputs = tf.concat(axis=1, values=[z, y])
G_h1 = tf.nn.relu(tf.matmul(inputs, G_W1) + G_b1)
G_h2 = tf.nn.relu(tf.matmul(G_h1, G_W2) + G_b2)
G_log_prob = tf.matmul(G_h2, G_W3) + G_b3
G_prob = tf.nn.sigmoid(G_log_prob)
return G_log_prob, G_prob
def sample_Z(m, n):
return np.random.uniform(-1., 1., size=[m, n])
###################################################
# Build CGAN Architecture
###################################################
print("Training CGAN...")
reg_dec = tf.keras.regularizers.l2(l2_norm_cgan)
G_sample, G_sample_sigmoid = generator(Z, y)
D_real, D_logit_real = discriminator(X, y)
D_fake, D_logit_fake = discriminator(G_sample, y)
D_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_logit_real, labels=tf.ones_like(D_logit_real)))
D_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_logit_fake, labels=tf.zeros_like(D_logit_fake)))
D_loss = D_loss_real + D_loss_fake
G_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_logit_fake, labels=tf.ones_like(D_logit_fake)))
D_solver = tf.train.AdamOptimizer(learning_rate=learning_rate_cgan).minimize(D_loss, var_list=theta_D)
G_solver = tf.train.AdamOptimizer(learning_rate=learning_rate_cgan).minimize(G_loss, var_list=theta_G)
train_side = side_data.values
train_latent = latent
train = tf.data.Dataset.from_tensor_slices((train_latent, train_side))
train = train.batch(1024)
train = train.shuffle(1000000)
train_iterator = train.make_initializable_iterator()
train_next_element = train_iterator.get_next()
sess = tf.Session()
sess.run(tf.global_variables_initializer())
for it in range(epoch_cgan):
sess.run(train_iterator.initializer)
batches = 0
total_g_loss = 0
total_d_loss = 0
while True:
try:
X_mb, y_mb = sess.run(train_next_element)
noise = np.random.normal(loc=0,scale=0.02,size=y_mb.shape[0]*y_mb.shape[1])
y_mb = np.clip(y_mb + noise.reshape(y_mb.shape[0], y_mb.shape[1]),0,1)
sums = np.sum(y_mb,1)
y_mb = y_mb/sums.reshape(-1,1)
batches += 1
Z_sample = sample_Z(X_mb.shape[0], Z_dim)
_, D_loss_curr = sess.run([D_solver, D_loss], feed_dict={X: X_mb, Z: Z_sample, y:y_mb})
_, G_loss_curr, samp = sess.run([G_solver, G_loss, G_sample], feed_dict={Z: Z_sample, y:y_mb})
except tf.errors.OutOfRangeError:
print(" %05d\tGen Loss: %.3f\tDisc Loss: %.3f" % (it, G_loss_curr, D_loss_curr), end="\r")
break
z_sample = sample_Z(side_data.values.shape[0], Z_dim)
gen_data = sess.run(G_sample, feed_dict={Z: z_sample, y:side_data.values})
gen_latent_df = pd.DataFrame(data=gen_data, index=sample_list)
gen_latent_df.to_csv("observed_generated_latent_values.tsv", sep="\t")
scatter = plot_data(gen_latent_df, side_data)
plt.title("Generated Latent Data (" + str(plot_method) + ")")
plt.savefig(out_dir + "/generated_latent_scatter.png")
plt.clf()
gen_decoded = dec_model.predict(gen_data)
gen_decoded_df = pd.DataFrame(data=gen_decoded, index=sample_list, columns=gene_list)
scatter = plot_data(gen_decoded_df, side_data)
plt.title("Generated Decoded Data (" + str(plot_method) + ")")
plt.savefig(out_dir + "/generated_decoded_scatter.png")
plt.clf()
###################################################
# Generate pseudocell trajectories
###################################################
print("\nGenerating data...")
states = side_data.columns.values
for c1_state in start_states:
for c2_state in end_states:
y_sample = []
for i in range(0,num_steps + 1):
for j in range(num_cells_gen):
c1 = 1 - i/num_steps
c2 = i/num_steps
c1_idx = np.where(states==c1_state)
c2_idx = np.where(states==c2_state)
y_state = np.zeros(len(states))
y_state[c1_idx] = c1
y_state[c2_idx] = c2
y_sample.append(y_state)
y_sample = np.array(y_sample)
z_sample = sample_Z(y_sample.shape[0], Z_dim)
gen_data = sess.run(G_sample, feed_dict={Z: z_sample, y:y_sample})
data_df = pd.DataFrame(gen_data, index=range((num_steps + 1) * num_cells_gen))
y_df = pd.DataFrame(y_sample, index=range((num_steps + 1) * num_cells_gen), columns=states)
gen_dir = out_dir + "/" + c1_state + "_" + c2_state
try:
os.mkdir(gen_dir)
except:
pass
data_df.to_csv(gen_dir + "/generated_latent_data.tsv", sep="\t")
y_df.to_csv(gen_dir + "/generated_side_data.tsv", sep="\t")
dec_df = pd.DataFrame(dec_model.predict(gen_data), index=range((num_steps + 1) * num_cells_gen), columns=rna_data.columns)
dec_df.to_csv(gen_dir + "/generated_data.tsv", sep="\t")
g_idx = 0
full_genes_to_plot = np.concatenate((genes_to_plot, c1_state, c2_state), axis=None)
gene_plot_df = pd.DataFrame(data=np.zeros(((num_steps + 1) * num_cells_gen * len(full_genes_to_plot), 3)),
index=range((num_steps + 1) * num_cells_gen * len(full_genes_to_plot)),
columns=["State", "Gene", "Value"])
for g in full_genes_to_plot:
if g in gene_list:
offset = (num_steps + 1) * num_cells_gen * g_idx
for k in range((num_steps + 1) * num_cells_gen):
state_percent = y_df.iloc[k][c2_state]
gene_value = dec_df.iloc[k][g]
gene_plot_df.iloc[offset + k] = [state_percent, g, gene_value]
g_idx += 1
sns.lineplot(x="State", y="Value", hue="Gene", data=gene_plot_df)
plt.title(c1_state + " to " + c2_state)
plt.xlabel("% " + c2_state)
plt.savefig(gen_dir + "/genes.png")
plt.clf()