app.py

from Bio import PDB
from transformers import AutoTokenizer, AutoModelForCausalLM
from rdkit import Chem
import selfies as sf
import torch
import time
import re
import io
import gradio as gr

torch.manual_seed(int(time.time()))
if torch.cuda.is_available():
    torch.cuda.manual_seed_all(int(time.time()))

model_name = "ncfrey/ChemGPT-1.2B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

def load_pdb(file_obj):
    parser = PDB.PDBParser(QUIET=True)
    structure = parser.get_structure('protein', file_obj)
    return structure

def clean_and_decode_selfies(raw_output):
    tokens = re.findall(r'\[[^\[\]]+\]', raw_output)
    valid_tokens = [t for t in tokens if all(x not in t for x in ['Branch', 'Ring', 'expl'])]
    cleaned_selfies = ''.join(valid_tokens)
    try:
        smiles = sf.decoder(cleaned_selfies)
        mol = Chem.MolFromSmiles(smiles)
        if mol:
            return Chem.MolToSmiles(mol)
    except:
        return None

def generate_multiple_valid_smiles(prompt, n=10, max_length=100):
    valid_smiles = set()
    tries = 0
    while len(valid_smiles) < n and tries < n * 5:
        inputs = tokenizer(prompt, return_tensors="pt")
        outputs = model.generate(
            **inputs,
            max_length=max_length,
            do_sample=True,
            temperature=1.0,
            top_k=100,
            pad_token_id=tokenizer.eos_token_id
        )
        selfies_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
        smiles = clean_and_decode_selfies(selfies_output)
        if smiles:
            valid_smiles.add(smiles)
        tries += 1
    return list(valid_smiles)

def generate_drugs_from_pdb(pdb_file):
    try:
        with open(pdb_file.name, 'r') as f:
            pdb_str = f.read()
        pdb_io = io.StringIO(pdb_str)
        load_pdb(pdb_io)

        prompt = "Generate a molecule in SELFIES that binds to the mutated KRAS protein"
        smiles_list = generate_multiple_valid_smiles(prompt, n=10)

        if not smiles_list:
            return "❌ لم يتم توليد أي SMILES صالحة", ""

        smiles_text = "\n".join(smiles_list)
        return "✅ تم توليد المركبات بنجاح", smiles_text

    except Exception as e:
        return f"❌ خطأ: {str(e)}", ""

with gr.Blocks() as demo:
    gr.Markdown("# 🧬 توليد مركبات دوائية من ملف PDB باستخدام ChemGPT")
    with gr.Row():
        pdb_input = gr.File(label="📁 ارفع ملف PDB")
        run_btn = gr.Button("🚀 توليد SMILES")
    status = gr.Textbox(label="📢 الحالة")
    smiles_output = gr.Textbox(label="📄 المركبات (SMILES)", lines=10)
    run_btn.click(fn=generate_drugs_from_pdb, inputs=pdb_input, outputs=[status, smiles_output])

demo.launch()