/
plotting.py
219 lines (172 loc) · 9.86 KB
/
plotting.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
from matplotlib import pyplot as plt
import torch
import numpy as np
from mpl_toolkits.axes_grid1 import ImageGrid
import matplotlib.colors as colors
#from scipy.stats.stats import pearsonr
import plotly.graph_objects as go
import seaborn as sns
import pandas as pd
ez = torch.load('RUN_60_6_LayerDepth/expected_z.pt')
gamma = torch.load('RUN_60_6_LayerDepth/gamma.pt')
ell = torch.load('RUN_60_6_LayerDepth/ell.pt')
phi = torch.load('/home/ludeep/Desktop/power-law/prme/RUN_60_6_LayerDepth/phi.pt')
vocab_data = np.load('patient_vocab_all.npz')
vocab = vocab_data[vocab_data.files[0]]
del vocab_data
sumez = torch.sum(ez, dim=2)
topics = ez.shape[-1]
# only read in patient barcode, first line is mutation names
mutation_df = pd.read_csv('/home/ludeep/Desktop/power-law/prme/pat_mut_mat_all.csv',encoding='utf-8',usecols=[0],header=None,skiprows=[0])
# skipping nan values
tcga_tss_codes = pd.read_csv('/home/ludeep/Desktop/power-law/mutTCGA/tcga_codes.csv',encoding='utf-8',skiprows=[6,201,550])
M = torch.load('M_all.pt')
def plotPaintbox(ez, vocab, gamma):
# Plotting Strength via paintboxes
# https://stackoverflow.com/questions/19390320/scatterplot-contours-in-matplotlib
disp_y_array = torch.linspace( 0, 100,steps=100, dtype=torch.float)
disp_x_array = torch.linspace( 0, 100,steps=100, dtype=torch.float)
sumez = torch.sum(ez, dim=2)
topics = ez.shape[-1]
fig = plt.figure(figsize=(20., 20.))
grid = ImageGrid(fig, 111, # similar to subplot(111)
nrows_ncols=(6, 10), # creates 2x2 grid of axes
axes_pad=0.1, # pad between axes in inch.
)
for k, ax in enumerate(grid):
eznorm = (ez[:,:,k]/sumez).detach().cpu().numpy()
ax.pcolormesh(disp_x_array,disp_y_array,eznorm,vmin = -1., vmax = 1., cmap='jet')
top3_words = torch.sort(gamma[k,:], descending=True)[1][:3]
ax.annotate(s=vocab[top3_words.cpu()][0], xy=(.1,.1), xycoords='axes fraction', color="r")
ax.annotate(s=vocab[top3_words.cpu()][1], xy=(.1,.25), xycoords='axes fraction', color="r")
ax.annotate(s=vocab[top3_words.cpu()][2], xy=(.1,.4), xycoords='axes fraction', color="r")
ax.set_axis_off()
plt.savefig('topics_3.svg')
def plotTopWords(gamma, vocab, numTopics=30):
assert numTopics < gamma.shape[0], "numTopics shown must be less than or equal to number of total topics (gamma.shape[0])"
top10_words = []
#fig, ax = plt.subplots()
#fig.patch.set_visible(False)
#ax.axis('off')
#ax.axis('tight')
rows = []
for k in range(0,numTopics):
top10_words.append(list(vocab[(torch.sort(gamma[k,:], descending=True)[1][:10]).cpu()]))
rows.append('Topic {}'.format(k))
#ax.table(cellText=top10_words, rowLabels=rows, loc='center', fontsize=20)
#fig.tight_layout()
#plt.show()
fig = go.Figure(data=[go.Table(cells=dict(values=[np.array(rows).T.tolist(),
*(np.array(top10_words).T.tolist())], line_color='darkslategray',
fill_color='lightcyan', align='left'))])
fig.layout['template']['data']['table'][0]['header']['fill']['color']='rgba(0,0,0,0)'
fig.show()
#fig.write_image("top_words.jpeg")
def display_topics(numTopics, gamma, ell, vocab, top_n_words=10, top_n_similar_topics=5):
for k in range(numTopics):
topn_words = torch.sort(gamma[k,:],
descending=True)[1][:top_n_words]
topk_similar_topics = torch.sort(torch.norm(ell[k:(k+1),:].repeat(numTopics,1)-ell,
dim=1))[1][1:top_n_similar_topics+1]
print('Factor{}: Most similar to factor {}'.format(
k, topk_similar_topics.tolist()))
print(vocab[topn_words.cpu()])
def display_topics_patient_type(phi,cancer_type,patient_list,tcga_tss_codes,numTopics):
tss_codes = tcga_tss_codes[tcga_tss_codes['ABV'].str.contains(cancer_type)]['TSS'].tolist()
# process codes
for i in range(len(tss_codes)):
if len(tss_codes[i]) == 1:
tss_codes[i] = '-0' + tss_codes[i] + '-'
else:
tss_codes[i] = '-' + tss_codes[i] + '-'
# contains the index of barcodes that match with the patient sample of the specific cancer type
index_of_barcodes = patient_list[patient_list[0].str.contains('|'.join(tss_codes))].index.tolist()
phi2 = np.asarray(phi)[index_of_barcodes] # is of number of patients matching cancer type
best_topic_in_patient = []
best_topic_all_patient = np.zeros((numTopics,1))
for k in range(len(phi2)):
phi3 = phi2[k].detach().cpu() # shape num_of_topics x number of mutations in patient k
temp = torch.sort(phi3,dim=0,descending=True)[1].numpy()
temp_unique, temp_counts = np.unique(temp[0],return_counts=True)
for i, unq in enumerate(temp_unique):
best_topic_all_patient[unq] += temp_counts[i]
#len_of_phi3 = phi3.shape[-1]
#for j in range(len_of_phi3):
# # get the best topic for each mutation
# best_topic_in_patient.append(torch.sort(phi3[j],dim=0,descending=True)[1][0].numpy())
df2 = pd.DataFrame(np.log10(best_topic_all_patient),index=np.arange(0,60),columns=[cancer_type])
plt.figure(figsize=(20,20))
sns.barplot(x=np.arange(0,60),y=cancer_type,data=df2)
#plt.savefig('{}_factors.png'.format(cancer_type))
return df2
def norm_topics(topTopics, numTopics, gamma, ell, vocab, top_n_words=10):
bestTopics = topTopics.sort_values(ascending=False,by=topTopics.columns.values[0]).index.values[:30].tolist() # Get 10 best topics
topk_least_similar_topics = torch.ones(len(bestTopics),10)
distx = []
for k, topic in enumerate(bestTopics):
topk_least_similar_topics[k] = torch.sort(torch.norm(ell[topic:(topic+1),:].repeat(numTopics,1)-ell,dim=1),descending=True)[1][1:11]
print('Factor {}: Least similar to Factor {}'.format(bestTopics[k], topk_least_similar_topics[k].tolist()))
tempidx = int(topk_least_similar_topics[k][0])
#topn_words = torch.sort(gamma[topic,:],descending=True)[1][:top_n_words]
topn_words = torch.sort(gamma[topic,:],descending=True)[1]
topn_words_least_topic = torch.sort(gamma[tempidx,:],descending=True)[1][:top_n_words]
print('-' * 89)
print('Factor {} has these top mutations: '.format(bestTopics[k]))
vocabx = vocab[topn_words[:20].cpu()]
print(vocabx)
tempEGFR = np.where(vocab[topn_words.cpu()]=='ARID1A')
print('EGFR occurs at location {}'.format(tempEGFR))
tempKRAS = np.where(vocab[topn_words.cpu()]=='SOX9')
print('KRAS occurs at location {}'.format(tempKRAS))
distx.append(np.absolute(tempEGFR[0]-tempKRAS[0]))
print('-' * 89)
#print('Factor {} has these top mutations: '.format(topk_least_similar_topics[k][0]))
#print(vocab[topn_words_least_topic.cpu()])
print (np.mean(distx))
def display_topics_TML(phi,M,cancer_type,patient_list,tcga_tss_codes,numTopics):
tss_codes = tcga_tss_codes[tcga_tss_codes['ABV'].str.contains(cancer_type)]['TSS'].tolist()
# process codes
for i in range(len(tss_codes)):
if len(tss_codes[i]) == 1:
tss_codes[i] = '-0' + tss_codes[i] + '-'
else:
tss_codes[i] = '-' + tss_codes[i] + '-'
index_of_barcodes = patient_list[patient_list[0].str.contains('|'.join(tss_codes))].index.tolist()
#tempx = index_of_barcodes[:20]
phi2 = np.asarray(phi)[index_of_barcodes]
M_2 = np.asarray(M)[index_of_barcodes]
#best_topic = np.zeros((numTopics,1)
best_topic = pd.DataFrame(data=np.zeros((len(index_of_barcodes),numTopics+1)),index=None,columns=[*np.arange(0,60), 'TML'])
for k in range(len(phi2)):
phi3 = phi2[k].detach().cpu() # shape num_of_topics x number of mutations in patient k
temp = torch.sort(phi3,dim=0,descending=True)[1].numpy() # top 5 factors
best_topic.iloc[k,np.ravel(np.unique(temp[0]))] = 1
best_topic['TML'].iloc[k] = np.log10(M_2[k])
#print("With TML: {}".format(M_2[k]))
#print(best_topic)
best_topic2 = pd.melt(best_topic, id_vars=['TML'])
#plt.yscale('log')
#plt.tight_layout()
#plt.figure(figsize=(20,20))
ax = sns.catplot(x='variable', y='TML', jitter=False, data=best_topic2.query("value != 0"), height=5, aspect=2)
ax.set(xlabel='Factor Number', ylabel='Log Scale Tumor Mutation Load', title='Mutually Exclusivity of TML and Mutations in COAD')
ax.savefig('COAD_TMLvsTopics.pdf')
#plotTopWords(gamma.detach(), vocab)
#display_topics(topics, gamma.detach(), ell.detach(), vocab, top_n_words=10, top_n_similar_topics=5)
#dfluad = display_topics_patient_type(phi,'LUAD',mutation_df,tcga_tss_codes,topics)
#norm_topics(dfluad, topics, gamma.detach(), ell.detach().cpu(), vocab)
dfcoad = display_topics_patient_type(phi,'COAD',mutation_df,tcga_tss_codes,topics)
norm_topics(dfcoad, topics, gamma.detach(), ell.detach().cpu(), vocab)
#display_topics_TML(phi,M,'COAD',mutation_df,tcga_tss_codes,topics)
'''
dfcoad = display_topics_patient_type(phi,'COAD',mutation_df,tcga_tss_codes,topics)
dfread = display_topics_patient_type(phi,'READ',mutation_df,tcga_tss_codes,topics)
dfbrca = display_topics_patient_type(phi,'BRCA',mutation_df,tcga_tss_codes,topics)
dfluad = display_topics_patient_type(phi,'LUAD',mutation_df,tcga_tss_codes,topics)
finaldf = pd.concat([dfluad, dfcoad, dfread, dfbrca], axis=1)
finaldf['index_col'] = finaldf.index
df3 = pd.melt(finaldf,id_vars=['index_col'])
ax = sns.catplot(x='index_col',y='value',hue='variable',kind='bar',data=df3.query("value > 0.1"), height=5,aspect=2)
ax.set(xlabel='Factor Number', ylabel='Log Scale Count of Factors', title='Number of Factors in a Cancer-Type')
ax.savefig('combined_cancer_box.pdf')
'''