Upload README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,95 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
license: mit
|
| 5 |
+
tags:
|
| 6 |
+
- genomics
|
| 7 |
+
- dna
|
| 8 |
+
- chicken
|
| 9 |
+
- poultry
|
| 10 |
+
- caduceus
|
| 11 |
+
- mamba
|
| 12 |
+
- biology
|
| 13 |
+
- bioinformatics
|
| 14 |
+
- gallus-gallus
|
| 15 |
+
datasets:
|
| 16 |
+
- custom
|
| 17 |
+
pipeline_tag: feature-extraction
|
| 18 |
+
base_model: kuleshov-group/caduceus-ph_seqlen-131k_d_model-256_n_layer-4
|
| 19 |
+
---
|
| 20 |
+
|
| 21 |
+
# π PoultryCaduceus
|
| 22 |
+
|
| 23 |
+
**A Bidirectional DNA Language Model for Chicken Genome**
|
| 24 |
+
|
| 25 |
+
PoultryCaduceus is the first DNA foundation model specifically pre-trained on the chicken (*Gallus gallus*) genome, based on the [Caduceus](https://github.com/kuleshov-group/caduceus) architecture.
|
| 26 |
+
|
| 27 |
+
## Model Description
|
| 28 |
+
|
| 29 |
+
| Feature | Value |
|
| 30 |
+
|---------|-------|
|
| 31 |
+
| **Base Model** | caduceus-ph (4-layer) |
|
| 32 |
+
| **Pre-training Genome** | GRCg6a (galGal6) |
|
| 33 |
+
| **Sequence Length** | 65,536 bp |
|
| 34 |
+
| **Hidden Dimension** | 256 |
|
| 35 |
+
| **Layers** | 4 |
|
| 36 |
+
| **Vocab Size** | 16 |
|
| 37 |
+
| **Training Steps** | 10,000 |
|
| 38 |
+
|
| 39 |
+
## Usage
|
| 40 |
+
|
| 41 |
+
```python
|
| 42 |
+
from transformers import AutoModelForMaskedLM
|
| 43 |
+
|
| 44 |
+
# Load model
|
| 45 |
+
model = AutoModelForMaskedLM.from_pretrained(
|
| 46 |
+
"jamie0315/PoultryCaduceus",
|
| 47 |
+
subfolder="checkpoint-10000",
|
| 48 |
+
trust_remote_code=True
|
| 49 |
+
)
|
| 50 |
+
```
|
| 51 |
+
|
| 52 |
+
### Get Sequence Embeddings
|
| 53 |
+
|
| 54 |
+
```python
|
| 55 |
+
import torch
|
| 56 |
+
|
| 57 |
+
# DNA vocabulary
|
| 58 |
+
DNA_VOCAB = {'A': 7, 'C': 8, 'G': 9, 'T': 10, 'N': 5, '[MASK]': 4}
|
| 59 |
+
|
| 60 |
+
# Tokenize sequence
|
| 61 |
+
sequence = "ATGCGATCGATCGATCG"
|
| 62 |
+
input_ids = torch.tensor([[DNA_VOCAB.get(c, 5) for c in sequence]])
|
| 63 |
+
|
| 64 |
+
# Get embeddings
|
| 65 |
+
model.eval()
|
| 66 |
+
with torch.no_grad():
|
| 67 |
+
outputs = model(input_ids, output_hidden_states=True)
|
| 68 |
+
embeddings = outputs.hidden_states[-1] # (batch, seq_len, 256)
|
| 69 |
+
```
|
| 70 |
+
|
| 71 |
+
## Training Data
|
| 72 |
+
|
| 73 |
+
Pre-training data is available in `chicken_pretrain_data_GRCg6a/`:
|
| 74 |
+
- `train_65k.h5` - Training set (~58,000 sequences)
|
| 75 |
+
- `val_65k.h5` - Validation set (~1,200 sequences)
|
| 76 |
+
|
| 77 |
+
## Repository Structure
|
| 78 |
+
|
| 79 |
+
```
|
| 80 |
+
PoultryCaduceus/
|
| 81 |
+
βββ checkpoint-10000/ # Model weights
|
| 82 |
+
β βββ config.json
|
| 83 |
+
β βββ model.safetensors
|
| 84 |
+
βββ chicken_pretrain_data_GRCg6a/ # Pre-training data
|
| 85 |
+
βββ train_65k.h5
|
| 86 |
+
βββ val_65k.h5
|
| 87 |
+
```
|
| 88 |
+
|
| 89 |
+
## Links
|
| 90 |
+
|
| 91 |
+
- π¦ **GitHub**: [chengzhimin/PoultryCaduceus](https://github.com/chengzhimin/PoultryCaduceus)
|
| 92 |
+
|
| 93 |
+
## License
|
| 94 |
+
|
| 95 |
+
MIT License
|