Yiquan2/Flu_Foundation

🦠 Flu Virus Foundation Model

This is a foundation model trained on influenza virus sequences for predicting viral evolution, functional constraints, and potential antigenic changes.
It is designed to support research in influenza biology, vaccine design, and immunology.

---

📖 Model Details

• Model type: Transformer-based language model
• Architecture: GPT-2 style
• Framework: 🤗 Transformers (PyTorch)
• Files included: model weights (model.safetensors), tokenizer, config files
• Trained on: pretrained on GISAID and NCBI influenza A sequences; finetuning on Noncoding region sequence completion

---

🚀 Simple Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Model and tokenizer from Hugging Face Hub
model_name = "Yiquan2/Flu_Foundation"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Example input sequence (DNA/protein)
sequence = "ATGAATCCAAACCAGAAAATAATAACCATTGGCTCTGTT"

# Tokenize input
inputs = tokenizer(sequence, return_tensors="pt")

# Generate output probabilities or predictions
with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits  # shape: [batch_size, seq_len, vocab_size]

# Optional: compute probabilities
probs = torch.softmax(logits, dim=-1)

print(probs)

🧪 Example: Mutation Effect Prediction with Flu Foundation Model

python mutation_prediction.py \
    --csv DMS_NA_data/Mos99_fit.csv \
    --fasta Mos99_nucleotide.fasta \
    --model Yiquan2/Flu_Foundation \
    --output results.csv