app.py

import gradio as gr
import numpy as np
import plotly.graph_objects as go

# Final model coefficients from Elastic Net
INTERCEPT = 2.384
COEFS = {
    "Age": -0.112,
    "C3": -0.327,
    "C4": -0.053,
    "C3/C4": -0.166,
    "Ln_C3C4": 0.135,
    "Ln_C4C3": -0.135
}
THRESHOLD = 0.461

# ------------------------
# Prediction Function
# ------------------------
def predict(age, c3, c4):
    if c4 == 0 or c3 == 0:
        return "Invalid input", None, None, "C3 and C4 must be > 0", ""

    c3_c4 = c3 / c4
    ln_c3c4 = np.log(c3_c4)
    ln_c4c3 = np.log(c4 / c3)

    logit = INTERCEPT + (
        COEFS["Age"] * age +
        COEFS["C3"] * c3 +
        COEFS["C4"] * c4 +
        COEFS["C3/C4"] * c3_c4 +
        COEFS["Ln_C3C4"] * ln_c3c4 +
        COEFS["Ln_C4C3"] * ln_c4c3
    )
    prob = 1 / (1 + np.exp(-logit))

    label = "Likely To Develop Renal Complications" if prob > THRESHOLD else "Not Likely To Develop Renal Complications"
    confidence = f"Model confidence: {round(prob * 100 if prob > THRESHOLD else (1 - prob) * 100)}%"
    interpretation = (
        "This result indicates a higher-than-threshold probability of developing renal complications. "
        "Further clinical evaluation and monitoring is advised."
        if prob > THRESHOLD else
        "The probability of abnormal renal function appears low. Routine follow-up is sufficient unless other risk factors exist."
    )

    return label, create_logit_meter(prob), create_prob_gauge(prob), confidence, interpretation

# ------------------------
# Linear Logit Meter
# ------------------------
def create_logit_meter(prob):
    fig = go.Figure(go.Indicator(
        mode="gauge+number+delta",
        value=prob,
        number={'valueformat': ".3f", 'font': {'size': 36, 'color': "black"}},
        delta={'reference': THRESHOLD, 'increasing': {'color': "red"}, 'decreasing': {'color': "green"}},
        gauge={
            'shape': "bullet",
            'axis': {
                'range': [0, 1],
                'tickwidth': 2,
                'tickcolor': "#666",
                'tickfont': {'size': 16}
            },
            'bar': {'color': "blue", 'thickness': 0.4},
            'steps': [
                {'range': [0, THRESHOLD], 'color': "#ccffcc"},
                {'range': [THRESHOLD, 1], 'color': "#ad3d46"}
            ],
            'threshold': {
                'line': {'color': "black", 'width': 4},
                'thickness': 0.8,
                'value': THRESHOLD
            },
        },
        title={
            "text": f"Logit Risk Probability",
            "font": {'size': 18}
        },
        domain={'x': [0, 1], 'y': [0, 1]}
    ))

    fig.update_layout(
        height=240,
        paper_bgcolor="white",
        font=dict(family="Arial", size=16, color="black"),
        margin=dict(t=40, b=30, l=10, r=10)
    )

    return fig


# ------------------------
# Dynamic Probability Gauge
# ------------------------
def create_prob_gauge(prob):
    if prob > THRESHOLD:
        gauge_value = prob
        title = "Probability of Abnormal Renal Function"
        steps = [
            {'range': [0, 0.33], 'color': "lightgreen"},
            {'range': [0.33, 0.66], 'color': "orange"},
            {'range': [0.66, 1], 'color': "red"}
        ]
    else:
        gauge_value = 1 - prob
        title = "Probability of Normal Renal Function"
        steps = [
            {'range': [0, 0.33], 'color': "red"},
            {'range': [0.33, 0.66], 'color': "orange"},
            {'range': [0.66, 1], 'color': "lightgreen"}
        ]

    fig = go.Figure(go.Indicator(
        mode="gauge+number",
        value=gauge_value,
        number={'valueformat': ".0%"},
        gauge={
            'axis': {'range': [0, 1]},
            'bar': {'color': "black"},
            'steps': steps,
        },
        title={"text": title}
    ))
    fig.update_layout(height=500, margin=dict(t=10, b=10, l=10, r=10))
    return fig

# ------------------------
# Gradio Interface
# ------------------------
demo = gr.Interface(
    fn=predict,
    inputs=[
        gr.Slider(5, 18, step=0.5, label="Age (Years)"),
        gr.Number(label="C3 Level (mg/dL)"),
        gr.Number(label="C4 Level (mg/dL)")
    ],
    outputs=[
        gr.Text(label="Screening Verdict"),
        gr.Plot(label="Logit Risk Meter"),
        gr.Plot(label="Risk Probability Gauge"),
        gr.Text(label="Model Confidence"),
        gr.Textbox(label="Interpretation", lines=3)
    ],
    title="🧪 Renal Complication Prediction Tool In Cases With Paediatric SLE",
    description="Enter Age, C3, and C4 levels to Predict Probability of Future Renal Complication based on a validated dataset of Paeditric SLE dataset by robust Elastic Net Regression."
)

demo.launch()