gradio_app.py

# gradio_app.py
# Gradio rewrite of the original Streamlit app for TFBS prediction

import os
import gradio as gr
import numpy as np
from utils import dnaseq_features
from keras.models import load_model

# Load model once at startup (make sure best_model.h5 is present in the repo)
model = load_model('best_model.h5')


def predict(dna_seq: str):
    """Take a DNA sequence string, compute features, run the model and return a Markdown summary."""
    dna_seq = (dna_seq or "").strip()
    if not dna_seq:
        return "**Please provide an input DNA sequence.**"

    # compute features using user's util function
    try:
        dna_ohe_feat, ds_index, ds_val = dnaseq_features(seq=dna_seq)
    except Exception as e:
        return f"Error while computing features: {e}"

    # predict
    try:
        predicted = model.predict(dna_ohe_feat)
    except Exception as e:
        return f"Error during model prediction: {e}"

    # build output as Markdown list with probabilities/scores
    lines = []
    for i, j in zip(ds_val, predicted):
        try:
            # if prediction is a vector like [p0, p1]
            if hasattr(j, "__len__") and len(j) >= 2:
                prob = float(j[1])
                label = "**TFBS found ✅**" if np.argmax(j) == 1 else "**TFBS not found ❌**"
                lines.append(f"- `{i}` — {label} (probability: {prob:.4f})")
            else:
                # single scalar output
                score = float(j)
                label = "**TFBS found ✅**" if score > 0.5 else "**TFBS not found ❌**"
                lines.append(f"- `{i}` — {label} (score: {score:.4f})")
        except Exception:
            # fallback
            lines.append(f"- `{i}` — prediction: {j}")

    if not lines:
        return "No predictions were produced. Check the input or the model."

    return "\n".join(lines)


# Gradio UI
title = "Simple Model Serving Web App for TFBS prediction"
description = "Get TFBS predictions from the latest model. Paste a DNA sequence and click **Make Prediction**."

with gr.Blocks(theme=None) as demo:
    gr.Markdown(f"# {title}")
    gr.Markdown(description)

    with gr.Row():
        seq_input = gr.Textbox(lines=6, placeholder="ATAGAGAC...", label="Input DNA sequence")

    with gr.Row():
        predict_btn = gr.Button("Make Prediction")

    output = gr.Markdown()

    predict_btn.click(fn=predict, inputs=seq_input, outputs=output)


if __name__ == "__main__":
    port = int(os.environ.get("PORT", 7860))
    demo.launch(server_name="0.0.0.0", server_port=port)