app.py

import torch
import joblib
import gradio as gr
import numpy as np
import pandas as pd
from transformers import T5Tokenizer, T5EncoderModel
import os

# --- Configuration ---
DEVICE = torch.device("cpu")
MODEL_PATH = "dna_binding_DNAstackBP_model.pkl" 
SCALER_PATH = "dna_binding_scaler.pkl"

# --- Load DNAstackBP Assets ---
print("Initializing DNAstackBP Dashboard...")
tokenizer = T5Tokenizer.from_pretrained("Rostlab/prot_t5_xl_uniref50", do_lower_case=False)
model = T5EncoderModel.from_pretrained("Rostlab/prot_t5_xl_uniref50", torch_dtype=torch.float16).to(DEVICE).eval()
svm_model = joblib.load(MODEL_PATH)
scaler = joblib.load(SCALER_PATH)

def predict_and_report(sequence):
    if not sequence or len(sequence.strip()) < 5:
        return "Invalid Sequence", 0.0, None, "Please enter a valid protein sequence."
    
    clean_seq = sequence.upper().strip()
    seq_fmt = " ".join(list(clean_seq))
    ids = tokenizer(seq_fmt, return_tensors="pt", padding=True)
    
    with torch.no_grad():
        embedding = model(input_ids=ids['input_ids'].to(DEVICE), attention_mask=ids['attention_mask'].to(DEVICE))
    
    last_hidden = embedding.last_hidden_state.squeeze(0)
    mask = ids['attention_mask'].squeeze(0).unsqueeze(-1)
    mean_embedding = torch.sum(last_hidden * mask, dim=0) / torch.sum(mask)
    
    feat = scaler.transform(mean_embedding.cpu().numpy().reshape(1, -1))
    label = svm_model.predict(feat)[0]
    prob = svm_model.predict_proba(feat)[0]
    
    result_text = "DNA-Binding Protein" if label == 1 else "Non-Binding Protein"
    conf_score = float(prob[label])
    conf_pct = f"{conf_score:.2%}"

    report_file = "DNAstackBP_Analysis_Report.csv"
    report_df = pd.DataFrame([{
        "Serial No.": 1,
        "Tool": "DNAstackBP",
        "Input Sequence": clean_seq,
        "Sequence Length": len(clean_seq),
        "Result": result_text,
        "Confidence": conf_pct
    }])
    report_df.to_csv(report_file, index=False)
    
    status_msg = f"Analysis Complete! Confidence: {conf_pct}."
    
    return result_text, conf_score, report_file, status_msg

# --- Custom Styling ---
# We use a Slate/Blue theme for a professional "Lab" look
theme = gr.themes.Soft(
    primary_hue="cyan",
    secondary_hue="slate",
    neutral_hue="slate",
    font=[gr.themes.GoogleFont("Inter"), "ui-sans-serif", "system-ui", "sans-serif"],
).set(
    body_background_fill="*neutral_50",
    block_background_fill="white",
    block_border_width="1px",
    button_primary_background_fill="*primary_600",
    button_primary_background_fill_hover="*primary_500",
)

with gr.Blocks(theme=theme) as demo:
    
    # Realistic Header with DNA Graphic
    with gr.Row():
        gr.HTML(
            """
            <div style="display: flex; align-items: center; gap: 20px; padding: 20px; background: white; border-bottom: 3px solid #0891b2; border-radius: 8px;">
                <img src="https://cdn-icons-png.flaticon.com/512/3011/3011406.png" width="80" height="80" style="filter: drop-shadow(2px 4px 6px #ccc);">
                <div>
                    <h1 style="margin: 0; color: #0f172a; font-size: 2.5rem;">DNAstackBP</h1>
                    <p style="margin: 0; color: #64748b; font-size: 1.1rem;">High-Precision DNA-Binding Protein Prediction System</p>
                </div>
            </div>
            """
        )

    with gr.Tab("Analysis Dashboard"):
        with gr.Row():
            with gr.Column(scale=2):
                gr.Markdown("### 🧬 Input Sequence")
                input_text = gr.Textbox(
                    label="Paste Protein Sequence (Raw or FASTA)", 
                    lines=12, 
                    placeholder="Example: MASG...",
                )
                with gr.Row():
                    clear_btn = gr.Button("🗑️ Clear", variant="secondary")
                    submit_btn = gr.Button("🔍 Run Prediction", variant="primary")
            
            with gr.Column(scale=1):
                gr.Markdown("### 📝 Results")
                output_status = gr.Markdown("*System Ready*")
                output_text = gr.Label(label="Prediction Result")
                output_num = gr.Number(label="Confidence Level", interactive=False)
                output_file = gr.File(label="📥 Export CSV Report")

    with gr.Tab("About DNAstackBP"):
        gr.Markdown(
            """
            ### Technical Methodology
            **DNAstackBP** utilizes a multi-stage deep learning pipeline:
            
            1. **Embedding Stage:** Uses **ProtT5-XL-UniRef50** for high-dimensional feature extraction.
            2. **Baseline Dataset:** Generates probability values from curated baseline models.
            3. **Meta-Classification:** Final analysis via a **CNN-1D Meta-Model** architecture.
            
            ### Developer Profile
            **Md. Shafiqul Islam** **ID: 221-15-4656**
            
            *Bioinformatics Analysis Tool - v1.0*
            """
        )

    submit_btn.click(
        fn=predict_and_report, 
        inputs=input_text, 
        outputs=[output_text, output_num, output_file, output_status]
    )
    clear_btn.click(lambda: ["", None, None, "*System Ready*"], outputs=[input_text, output_text, output_file, output_status])

if __name__ == "__main__":
    demo.launch()