app.py

import os
import json
import streamlit as st
from transformers import AutoTokenizer, AutoModel
import torch
import numpy as np
import faiss

# Page configuration
st.set_page_config(page_title='KRISSBERT UMLS Linker', layout='wide')
st.title('🧬 KRISSBERT + UMLS Entity Linker (Local FAISS)')

# Paths & model name
METADATA_PATH = 'umls_metadata.json'
EMBED_PATH = 'umls_embeddings.npy'
INDEX_PATH = 'umls_index.faiss'
MODEL_NAME = 'microsoft/BiomedNLP-KRISSBERT-PubMed-UMLS-EL'

# 1️⃣ Load model & tokenizer
@st.cache_resource
def load_model():
    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
    model = AutoModel.from_pretrained(MODEL_NAME)
    model.eval()
    return tokenizer, model

tokenizer, model = load_model()

# 2️⃣ Load UMLS FAISS index & metadata
@st.cache_resource
def load_umls_index():
    meta = json.load(open(METADATA_PATH, 'r'))
    embeddings = np.load(EMBED_PATH)
    index = faiss.read_index(INDEX_PATH)
    return index, meta

faiss_index, umls_meta = load_umls_index()

# 3️⃣ Embed text (prefix underscores to avoid caching errors)
@st.cache_resource
def embed_text(text, _tokenizer, _model):
    inputs = _tokenizer(text, return_tensors='pt', truncation=True, padding=True)
    with torch.no_grad():
        outputs = _model(**inputs)
    emb = outputs.last_hidden_state[:, 0, :].squeeze().cpu().numpy()
    return emb / np.linalg.norm(emb)

# UI: examples + input
st.markdown('Enter a biomedical sentence to link entities via local UMLS FAISS index and KRISSBERT:')
examples = [
    'The patient was administered metformin for type 2 diabetes.',
    'ER crowding has become a widespread issue in hospitals.',
    'Tamoxifen is used in the treatment of ER-positive breast cancer.'
]
selected = st.selectbox('🔍 Example queries', ['Choose...'] + examples)
sentence = st.text_area('📝 Sentence:', value=(selected if selected != 'Choose...' else ''))

if st.button('🔗 Link Entities'):
    if not sentence.strip():
        st.warning('Please enter a sentence first.')
    else:
        with st.spinner('Embedding sentence and searching FAISS…'):
            sent_emb = embed_text(sentence, tokenizer, model).reshape(1, -1)
            distances, indices = faiss_index.search(sent_emb, 5)
            results = []
            for idx in indices[0]:
                entry = umls_meta.get(str(idx), {})
                results.append({
                    'cui': entry.get('cui', ''),
                    'name': entry.get('name', ''),
                    'definition': entry.get('definition', ''),
                    'source': entry.get('source', '')
                })
        # Display
        if results:
            st.success('Top UMLS candidates:')
            for item in results:
                st.markdown(f"**{item['name']}** (CUI: `{item['cui']}`)")
                if item['definition']:
                    st.markdown(f"> {item['definition']}\n")
                st.markdown(f"_Source: {item['source']}_\n---")
        else:
            st.info('No matches found in UMLS index.')