AmelieSchreiber
commited on
Commit
•
ebf710f
1
Parent(s):
a2760f2
Update README.md
Browse files
README.md
CHANGED
@@ -1,9 +1,81 @@
|
|
1 |
---
|
2 |
library_name: peft
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
## Training procedure
|
5 |
|
6 |
-
|
|
|
|
|
7 |
|
|
|
8 |
|
9 |
- PEFT 0.4.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
library_name: peft
|
3 |
+
license: mit
|
4 |
+
language:
|
5 |
+
- en
|
6 |
+
tags:
|
7 |
+
- transformers
|
8 |
+
- biology
|
9 |
+
- esm
|
10 |
+
- esm2
|
11 |
+
- protein
|
12 |
+
- protein language model
|
13 |
---
|
14 |
+
# ESM-2 RNA Binding Site LoRA
|
15 |
+
|
16 |
+
This is a Parameter Efficient Fine Tuning (PEFT) Low Rank Adaptation ([LoRA](https://huggingface.co/docs/peft/task_guides/token-classification-lora)) of
|
17 |
+
the [esm2_t12_35M_UR50D](https://huggingface.co/facebook/esm2_t12_35M_UR50D) model for the (binary) token classification task of
|
18 |
+
predicting RNA binding sites of proteins. The Github with the training script and conda env YAML can be
|
19 |
+
[found here](https://github.com/Amelie-Schreiber/esm2_LoRA_binding_sites/tree/main). You can also find a version of this model
|
20 |
+
that was fine-tuned without LoRA [here](https://huggingface.co/AmelieSchreiber/esm2_t6_8M_UR50D_rna_binding_site_predictor).
|
21 |
+
|
22 |
## Training procedure
|
23 |
|
24 |
+
This is a Low Rank Adaptation (LoRA) of `esm2_t6_8M_UR50D`,
|
25 |
+
trained on `166` protein sequences in the [RNA binding sites dataset](https://huggingface.co/datasets/AmelieSchreiber/data_of_protein-rna_binding_sites)
|
26 |
+
using a `75/25` train/test split. It achieves an evaluation loss of `0.18801096081733704`.
|
27 |
|
28 |
+
### Framework versions
|
29 |
|
30 |
- PEFT 0.4.0
|
31 |
+
|
32 |
+
## Using the Model
|
33 |
+
|
34 |
+
To use, try running:
|
35 |
+
```python
|
36 |
+
from transformers import AutoModelForTokenClassification, AutoTokenizer
|
37 |
+
from peft import PeftModel
|
38 |
+
import torch
|
39 |
+
|
40 |
+
# Path to the saved LoRA model
|
41 |
+
model_path = "AmelieSchreiber/esm2_t30_150M_LoRA_RNA_binding"
|
42 |
+
# ESM2 base model
|
43 |
+
base_model_path = "facebook/esm2_t30_150M_UR50D"
|
44 |
+
|
45 |
+
# Load the model
|
46 |
+
base_model = AutoModelForTokenClassification.from_pretrained(base_model_path)
|
47 |
+
loaded_model = PeftModel.from_pretrained(base_model, model_path)
|
48 |
+
|
49 |
+
# Ensure the model is in evaluation mode
|
50 |
+
loaded_model.eval()
|
51 |
+
|
52 |
+
# Load the tokenizer
|
53 |
+
loaded_tokenizer = AutoTokenizer.from_pretrained(base_model_path)
|
54 |
+
|
55 |
+
# Protein sequence for inference
|
56 |
+
protein_sequence = "MAVPETRPNHTIYINNLNEKIKKDELKKSLHAIFSRFGQILDILVSRSLKMRGQAFVIFKEVSSATNALRSMQGFPFYDKPMRIQYAKTDSDIIAKMKGT" # Replace with your actual sequence
|
57 |
+
|
58 |
+
# Tokenize the sequence
|
59 |
+
inputs = loaded_tokenizer(protein_sequence, return_tensors="pt", truncation=True, max_length=1024, padding='max_length')
|
60 |
+
|
61 |
+
# Run the model
|
62 |
+
with torch.no_grad():
|
63 |
+
logits = loaded_model(**inputs).logits
|
64 |
+
|
65 |
+
# Get predictions
|
66 |
+
tokens = loaded_tokenizer.convert_ids_to_tokens(inputs["input_ids"][0]) # Convert input ids back to tokens
|
67 |
+
predictions = torch.argmax(logits, dim=2)
|
68 |
+
|
69 |
+
# Define labels
|
70 |
+
id2label = {
|
71 |
+
0: "No binding site",
|
72 |
+
1: "Binding site"
|
73 |
+
}
|
74 |
+
|
75 |
+
# Print the predicted labels for each token
|
76 |
+
for token, prediction in zip(tokens, predictions[0].numpy()):
|
77 |
+
if token not in ['<pad>', '<cls>', '<eos>']:
|
78 |
+
print((token, id2label[prediction]))
|
79 |
+
|
80 |
+
```
|
81 |
+
|