Commit
•
fb2ce95
1
Parent(s):
4e2b988
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,82 @@
|
|
1 |
---
|
2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
library_name: transformers
|
6 |
+
tags:
|
7 |
+
- esm
|
8 |
+
- esm-2
|
9 |
+
- sequence classifier
|
10 |
+
- proteins
|
11 |
+
- protein language model
|
12 |
---
|
13 |
+
|
14 |
+
# ESM-2 Sequence Classifier
|
15 |
+
This is a small sequence classifier trained on synthetic data generate by GPT-4
|
16 |
+
which classifies sequences into three categories `enzymes`, `transport_proteins`, and `structural_proteins`.
|
17 |
+
To use the model, try running:
|
18 |
+
|
19 |
+
```
|
20 |
+
# Load the trained model and tokenizer
|
21 |
+
model = EsmForSequenceClassification.from_pretrained("./esm2_t6_8M_UR50D_sequence_classifier_v1")
|
22 |
+
tokenizer = AutoTokenizer.from_pretrained("facebook/esm2_t6_8M_UR50D")
|
23 |
+
|
24 |
+
# Suppose these are your new sequences that you want to classify
|
25 |
+
# Additional Family 0: Enzymes
|
26 |
+
new_sequences_0 = [
|
27 |
+
"ACGYLKTPKLADPPVLRGDSSVTKAICKPDPVLEK",
|
28 |
+
"GVALDECKALDYLPGKPLPMDGKVCQCGSKTPLRP",
|
29 |
+
"VLPGYTCGELDCKPGKPLPKCGADKTQVATPFLRG",
|
30 |
+
"TCGALVQYPSCADPPVLRGSDSSVKACKKLDPQDK",
|
31 |
+
"GALCEECKLCPGADYKPMDGDRLPAAATSKTRPVG",
|
32 |
+
"PAVDCKKALVYLPKPLPMDGKVCRGSKTPKTRPYG",
|
33 |
+
"VLGYTCGALDCKPGKPLPKCGADKTQVATPFLRGA",
|
34 |
+
"CGALVQYPSCADPPVLRGSDSSVKACKKLDPQDKT",
|
35 |
+
"ALCEECKLCPGADYKPMDGDRLPAAATSKTRPVGK",
|
36 |
+
"AVDCKKALVYLPKPLPMDGKVCRGSKTPKTRPYGR",
|
37 |
+
]
|
38 |
+
|
39 |
+
# Additional Family 1: Receptor Proteins
|
40 |
+
new_sequences_1 = [
|
41 |
+
"VGQRFYGGRQKNRHCELSPLPSACRGSVQGALYTD",
|
42 |
+
"KDQVLTVPTYACRCCPKMDSKGRVPSTLRVKSARS",
|
43 |
+
"PLAGVACGRGLDYRCPRKMVPGDLQVTPATQRPYG",
|
44 |
+
"CGVRLGYPGCADVPLRGRSSFAPRACMKKDPRVTR",
|
45 |
+
"RKGVAYLYECRKLRCRADYKPRGMDGRRLPKASTT",
|
46 |
+
"RPTGAVNCKQAKVYRGLPLPMMGKVPRVCRSRRPY",
|
47 |
+
"RLDGGYTCGQALDCKPGRKPPKMGCADLKSTVATP",
|
48 |
+
"LGTCRKLVRYPQCADPPVMGRSSFRPKACCRQDPV",
|
49 |
+
"RVGYAMCSPKLCSCRADYKPPMGDGDRLPKAATSK",
|
50 |
+
"QPKAVNCRKAMVYRPKPLPMDKGVPVCRSKRPRPY",
|
51 |
+
]
|
52 |
+
|
53 |
+
# Additional Family 2: Structural Proteins
|
54 |
+
new_sequences_2 = [
|
55 |
+
"VGKGFRYGSSQKRYLHCQKSALPPSCRRGKGQGSAT",
|
56 |
+
"KDPTVMTVGTYSCQCPKQDSRGSVQPTSRVKTSRSK",
|
57 |
+
"PLVGKACGRSSDYKCPGQMVSGGSKQTPASQRPSYD",
|
58 |
+
"CGKKLVGYPSSKADVPLQGRSSFSPKACKKDPQMTS",
|
59 |
+
"RKGVASLYCSSKLSCKAQYSKGMSDGRSPKASSTTS",
|
60 |
+
"RPKSAASCEQAKSYRSLSLPSMKGKVPSKCSRSKRP",
|
61 |
+
"RSDVSYTSCSQSKDCKPSKPPKMSGSKDSSTVATPS",
|
62 |
+
"LSTCSKKVAYPSSKADPPSSGRSSFSMKACKKQDPPV",
|
63 |
+
"RVGSASSEPKSSCSVQSYSKPSMSGDSSPKASSTSK",
|
64 |
+
"QPSASNCEKMSSYRPSLPSMSKGVPSSRSKSSPPYQ",
|
65 |
+
]
|
66 |
+
|
67 |
+
# Tokenize the sequences and convert to tensors
|
68 |
+
# Merge all sequences
|
69 |
+
new_sequences = new_sequences_0 + new_sequences_1 + new_sequences_2
|
70 |
+
inputs = tokenizer(new_sequences, return_tensors="pt", padding=True, truncation=True)
|
71 |
+
|
72 |
+
# Use the model to get the logits
|
73 |
+
with torch.no_grad():
|
74 |
+
logits = model(**inputs).logits
|
75 |
+
|
76 |
+
# Get the predicted class for each sequence
|
77 |
+
predicted_class_ids = torch.argmax(logits, dim=-1)
|
78 |
+
|
79 |
+
# Print the predicted class for each sequence
|
80 |
+
for sequence, predicted_class in zip(new_sequences, predicted_class_ids):
|
81 |
+
print(f"Sequence: {sequence}, Predicted class: {predicted_class.item()}")
|
82 |
+
```
|