Commit ·
887e87e
1
Parent(s): 13ec48b
update readme
Browse files
README.md
CHANGED
|
@@ -43,12 +43,21 @@ from transformers import AutoTokenizer, AutoModel
|
|
| 43 |
import torch
|
| 44 |
|
| 45 |
# proteins
|
| 46 |
-
tokenizer_prot = AutoTokenizer.from_pretrained(
|
| 47 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
|
| 49 |
proteins = ["MKTFFVLLL", "ABCDE"]
|
| 50 |
proteins = [" ".join(i) for i in proteins]
|
| 51 |
-
inputs_prot = tokenizer_prot(
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
|
| 53 |
with torch.no_grad():
|
| 54 |
outputs = encoder_prot(**inputs_prot)
|
|
@@ -56,12 +65,22 @@ with torch.no_grad():
|
|
| 56 |
mask = inputs_prot['attention_mask'].unsqueeze(-1).float()
|
| 57 |
prot_rep = (hidden * mask).sum(dim=1) / mask.sum(dim=1).clamp(min=1e-8)
|
| 58 |
|
|
|
|
| 59 |
# molecules
|
| 60 |
-
tokenizer_mol = AutoTokenizer.from_pretrained(
|
| 61 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
|
| 63 |
molecules = ["NCCc1nc(-c2ccccc2)cs1", "CC(=O)OCC(C)C"]
|
| 64 |
-
inputs_mol = tokenizer_mol(
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
|
| 66 |
with torch.no_grad():
|
| 67 |
outputs = encoder_mol(**inputs_mol)
|
|
@@ -69,8 +88,12 @@ with torch.no_grad():
|
|
| 69 |
mask = inputs_mol['attention_mask'].unsqueeze(-1).float()
|
| 70 |
mol_rep = (hidden * mask).sum(dim=1) / mask.sum(dim=1).clamp(min=1e-8)
|
| 71 |
|
|
|
|
| 72 |
# seqscreen
|
| 73 |
-
seqscreen = AutoModel.from_pretrained(
|
|
|
|
|
|
|
|
|
|
| 74 |
|
| 75 |
with torch.no_grad():
|
| 76 |
outputs = seqscreen(prot=prot_rep, mol=mol_rep)
|
|
|
|
| 43 |
import torch
|
| 44 |
|
| 45 |
# proteins
|
| 46 |
+
tokenizer_prot = AutoTokenizer.from_pretrained(
|
| 47 |
+
'facebook/esm2_t36_3B_UR50D'
|
| 48 |
+
)
|
| 49 |
+
encoder_prot = AutoModel.from_pretrained(
|
| 50 |
+
'facebook/esm2_t36_3B_UR50D'
|
| 51 |
+
).eval()
|
| 52 |
+
|
| 53 |
|
| 54 |
proteins = ["MKTFFVLLL", "ABCDE"]
|
| 55 |
proteins = [" ".join(i) for i in proteins]
|
| 56 |
+
inputs_prot = tokenizer_prot(
|
| 57 |
+
proteins,
|
| 58 |
+
return_tensors="pt",
|
| 59 |
+
padding=True
|
| 60 |
+
)
|
| 61 |
|
| 62 |
with torch.no_grad():
|
| 63 |
outputs = encoder_prot(**inputs_prot)
|
|
|
|
| 65 |
mask = inputs_prot['attention_mask'].unsqueeze(-1).float()
|
| 66 |
prot_rep = (hidden * mask).sum(dim=1) / mask.sum(dim=1).clamp(min=1e-8)
|
| 67 |
|
| 68 |
+
|
| 69 |
# molecules
|
| 70 |
+
tokenizer_mol = AutoTokenizer.from_pretrained(
|
| 71 |
+
'SaeedLab/MolDeBERTa-base-123M-mlc'
|
| 72 |
+
)
|
| 73 |
+
encoder_mol = AutoModel.from_pretrained(
|
| 74 |
+
'SaeedLab/MolDeBERTa-base-123M-mlc'
|
| 75 |
+
).eval()
|
| 76 |
+
|
| 77 |
|
| 78 |
molecules = ["NCCc1nc(-c2ccccc2)cs1", "CC(=O)OCC(C)C"]
|
| 79 |
+
inputs_mol = tokenizer_mol(
|
| 80 |
+
molecules,
|
| 81 |
+
return_tensors="pt",
|
| 82 |
+
padding=True
|
| 83 |
+
)
|
| 84 |
|
| 85 |
with torch.no_grad():
|
| 86 |
outputs = encoder_mol(**inputs_mol)
|
|
|
|
| 88 |
mask = inputs_mol['attention_mask'].unsqueeze(-1).float()
|
| 89 |
mol_rep = (hidden * mask).sum(dim=1) / mask.sum(dim=1).clamp(min=1e-8)
|
| 90 |
|
| 91 |
+
|
| 92 |
# seqscreen
|
| 93 |
+
seqscreen = AutoModel.from_pretrained(
|
| 94 |
+
'SaeedLab/SeqScreen-Frozen',
|
| 95 |
+
trust_remote_code=True
|
| 96 |
+
).eval()
|
| 97 |
|
| 98 |
with torch.no_grad():
|
| 99 |
outputs = seqscreen(prot=prot_rep, mol=mol_rep)
|