app.py

import numpy as np
import torch
import shap
from transformers import (
    pipeline,
    AutoTokenizer,
    AutoModelForSequenceClassification,
    AutoModelForTokenClassification
)
import gradio as gr

# 1) Device setup
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 2) Load ADR classifier model & tokenizer
model_name = "paragon-analytics/ADRv1"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name).to(device)

# 3) Build HF text-classification pipeline
pred_pipeline = pipeline(
    "text-classification",
    model=model,
    tokenizer=tokenizer,
    return_all_scores=True,
    device=0 if device.type == "cuda" else -1
)

# 4) Base predict_proba: List[str] → np.ndarray of shape (n_samples, n_classes)
def predict_proba(texts):
    if isinstance(texts, str):
        texts = [texts]
    results = pred_pipeline(texts)
    # results: List[List[{"label":…, "score":…}]]
    probs = np.array([[d["score"] for d in sample] for sample in results])
    return probs

# 5) SHAP-compatible wrapper: joins token lists back into strings
def predict_proba_shap(inputs):
    # inputs: List[str] or List[List[str]]
    texts = [
        " ".join(x) if isinstance(x, list) else x
        for x in inputs
    ]
    return predict_proba(texts)

# 6) Instantiate SHAP explainer with a Text masker
masker = shap.maskers.Text(tokenizer)
# Grab output class labels from a dummy sample
_example = pred_pipeline(["test"])[0]
class_labels = [d["label"] for d in _example]

explainer = shap.Explainer(
    predict_proba_shap,
    masker=masker,
    output_names=class_labels
)

# 7) Load biomedical NER model & pipeline
ner_model_name = "d4data/biomedical-ner-all"
ner_tokenizer = AutoTokenizer.from_pretrained(ner_model_name)
ner_model = AutoModelForTokenClassification.from_pretrained(ner_model_name).to(device)
ner_pipe = pipeline(
    "ner",
    model=ner_model,
    tokenizer=ner_tokenizer,
    aggregation_strategy="simple",
    device=0 if device.type == "cuda" else -1
)

# 8) Mapping for entity highlight colors
ENTITY_COLORS = {
    "Severity": "red",
    "Sign_symptom": "green",
    "Medication": "lightblue",
    "Age": "yellow",
    "Sex": "yellow",
    "Diagnostic_procedure": "gray",
    "Biological_structure": "silver"
}

# 9) Full predict + explain + NER function
def adr_predict(text: str):
    # a) Predict probabilities
    probs = predict_proba([text])[0]
    prob_dict = {label: float(probs[i]) for i, label in enumerate(class_labels)}

    # b) SHAP explanation → Matplotlib figure
    shap_values = explainer([text])
    fig = shap.plots.text(shap_values[0], display=False)

    # c) NER highlighting
    ents = ner_pipe(text)
    highlighted = ""
    last_idx = 0
    for ent in ents:
        start, end = ent["start"], ent["end"]
        word = ent["word"].replace("##", "")
        color = ENTITY_COLORS.get(ent["entity_group"], "lightgray")
        highlighted += (
            text[last_idx:start]
            + f"<mark style='background-color:{color};'>{word}</mark>"
        )
        last_idx = end
    highlighted += text[last_idx:]

    return prob_dict, fig, highlighted

# 10) Build Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## Welcome to **ADR Detector** 🪐")
    gr.Markdown(
        "Predicts the likelihood your text describes a **severe** vs. **non-severe** adverse reaction.  \n"
        "_(Not for medical or diagnostic use.)_"
    )

    txt = gr.Textbox(
        label="Enter Your Text Here:",
        lines=3,
        placeholder="Type a sentence about an adverse reaction…"
    )
    btn = gr.Button("Analyze")

    with gr.Row():
        label_out = gr.Label(label="Predicted Probabilities")
        shap_out = gr.Plot(label="SHAP Explanation")
        ner_out = gr.HTML(label="Biomedical Entities Highlighted")

    btn.click(
        fn=adr_predict,
        inputs=txt,
        outputs=[label_out, shap_out, ner_out]
    )

    gr.Examples(
        examples=[
            "A 35-year-old male experienced severe headache after taking Aspirin.",
            "A 35-year-old female had minor abdominal pain after Acetaminophen."
        ],
        inputs=txt,
        outputs=[label_out, shap_out, ner_out],
        fn=adr_predict,
        cache_examples=True
    )

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