Commit
·
a3b02b5
1
Parent(s):
794c624
update README
Browse files
README.md
CHANGED
@@ -1,22 +1,68 @@
|
|
1 |
|
|
|
|
|
2 |
|
3 |
```python
|
4 |
-
|
|
|
5 |
import re
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
tokenizer = PreTrainedTokenizerFast.from_pretrained("lixiangchun/transcriptome_iseeek_13millioncells_128tokens")
|
8 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
-
|
|
|
11 |
|
12 |
-
|
13 |
-
|
14 |
|
15 |
-
|
16 |
-
out = iseeek.bert(**batch)
|
17 |
|
18 |
-
|
19 |
-
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
```
|
22 |
|
|
|
1 |
|
2 |
+
# iSEEEK
|
3 |
+
A universal approach for integrating super large-scale single-cell transcriptomes by exploring gene rankings
|
4 |
|
5 |
```python
|
6 |
+
## An simple pipeline for single-cell analysis
|
7 |
+
import torch
|
8 |
import re
|
9 |
+
from tqdm import tqdm
|
10 |
+
import numpy as np
|
11 |
+
import scanpy as sc
|
12 |
+
from torch.utils.data import DataLoader, Dataset
|
13 |
+
from transformers import PreTrainedTokenizerFast, BertForMaskedLM
|
14 |
+
|
15 |
+
class LineDataset(Dataset):
|
16 |
+
def __init__(self, lines):
|
17 |
+
self.lines = lines
|
18 |
+
self.regex = re.compile(r'\-|\.')
|
19 |
+
def __getitem__(self, i):
|
20 |
+
return self.regex.sub('_', self.lines[i])
|
21 |
+
def __len__(self):
|
22 |
+
return len(self.lines)
|
23 |
+
|
24 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
25 |
+
torch.set_num_threads(2)
|
26 |
|
27 |
tokenizer = PreTrainedTokenizerFast.from_pretrained("lixiangchun/transcriptome_iseeek_13millioncells_128tokens")
|
28 |
+
model = BertForMaskedLM.from_pretrained("lixiangchun/transcriptome_iseeek_13millioncells_128tokens").bert
|
29 |
+
model = model.to(device)
|
30 |
+
model.eval()
|
31 |
+
|
32 |
+
|
33 |
+
text_file = "/mnt/ssd2/shenhr/BERT/bert_256/pbmc/deal/gene_rank_pmbc.txt"
|
34 |
+
labels = [s.strip() for s in open('/mnt/ssd2/shenhr/BERT/bert_256/pbmc/deal/labels.txt')]
|
35 |
+
labels = np.asarray(labels)
|
36 |
+
|
37 |
+
lines = [s.strip() for s in open(text_file)]
|
38 |
+
|
39 |
+
ds = LineDataset(lines)
|
40 |
+
dl = DataLoader(ds, batch_size=80)
|
41 |
+
|
42 |
+
features = []
|
43 |
+
|
44 |
+
for a in tqdm(dl, total=len(dl)):
|
45 |
+
batch = tokenizer(a, max_length=128, truncation=True,
|
46 |
+
padding=True, return_tensors="pt")
|
47 |
+
|
48 |
+
for k, v in batch.items():
|
49 |
+
batch[k] = v.to(device)
|
50 |
|
51 |
+
with torch.no_grad():
|
52 |
+
out = model(**batch)
|
53 |
|
54 |
+
f = out.last_hidden_state[:,0,:]
|
55 |
+
features.extend(f.tolist())
|
56 |
|
57 |
+
features = np.stack(features)
|
|
|
58 |
|
59 |
+
adata = sc.AnnData(features)
|
60 |
+
adata.obs['celltype'] = labels
|
61 |
+
adata.obs.celltype = adata.obs.celltype.astype("category")
|
62 |
+
sc.pp.neighbors(adata, use_rep='X')
|
63 |
+
sc.tl.umap(adata)
|
64 |
+
sc.tl.leiden(adata)
|
65 |
+
sc.pl.umap(adata, color=['celltype','leiden'],save= "UMAP")
|
66 |
|
67 |
```
|
68 |
|