app.py

import nltk
nltk.download('stopwords')
nltk.download('punkt')
import pandas as pd
import classify_abs
import extract_abs
pd.set_option('display.max_colwidth', None)
import streamlit as st
import spacy
import tensorflow as tf
import pickle
import plotly.graph_objects as go

#### LOGO ####
st.markdown('''<img src="https://huggingface.co/spaces/ncats/EpiPipeline4GARD/resolve/main/Logo_GARD_fullres.png" alt="NIH Genetic and Rare Diseases Information Center Logo" width=550>''',unsafe_allow_html=True)
#st.markdown("![National Center for Advancing Translational Sciences (NCATS) Logo](https://huggingface.co/spaces/ncats/EpiPipeline4GARD/resolve/main/NCATS_logo.png)")
#st.markdown('''<img src="https://huggingface.co/spaces/ncats/EpiPipeline4GARD/raw/main/NCATS_logo.svg" alt="National Center for Advancing Translational Sciences Logo" width="800" height="300">''',unsafe_allow_html=True)

#### TITLE ####
st.title("Epidemiology Extraction Pipeline for Rare Diseases")
#st.subheader("National Center for Advancing Translational Sciences (NIH/NCATS)") 

#### CHANGE SIDEBAR WIDTH ###
st.markdown(
    """
    <style>
    [data-testid="stSidebar"][aria-expanded="true"] > div:first-child {
        width: 250px;
    }
    [data-testid="stSidebar"][aria-expanded="false"] > div:first-child {
        width: 250px;
        margin-left: -350px;
    }
    </style>
    """,
    unsafe_allow_html=True,
)

#### DESCRIPTION ####
st.markdown("This application was built by the [National Center for Advancing Translational Sciences (NCATS)](https://ncats.nih.gov/) for the [National Institutes of Health (NIH)](https://www.nih.gov/) [Genetic and Rare Diseases Information Center](https://rarediseases.info.nih.gov/) to automatically search PubMed abstracts for rare disease epidemiology information.")

#### SIDEBAR WIDGETS ####

#max_results is Maximum number of PubMed ID's to retrieve BEFORE filtering
max_results = st.sidebar.number_input("Maximum number of articles to find in PubMed", min_value=1, max_value=None, value=50)

filtering = st.sidebar.radio("What type of filtering would you like?",('Strict', 'Lenient', 'None')).lower()

extract_diseases = st.sidebar.checkbox("Extract Rare Diseases", value=False)

#### MODEL LOADING ####

@st.experimental_singleton(show_spinner=False)
def load_models_experimental():
    classify_model_vars = classify_abs.init_classify_model()
    NER_pipeline, entity_classes = extract_abs.init_NER_pipeline()
    GARD_dict, max_length = extract_abs.load_GARD_diseases()
    return classify_model_vars, NER_pipeline, entity_classes, GARD_dict, max_length

#### DOWNLOAD FUNCTION ####

@st.cache
def convert_df(df):
    # IMPORTANT: Cache the conversion to prevent computation on every rerun
    return df.to_csv().encode('utf-8')
    
#### SANKEY FUNCTION ####

@st.cache(allow_output_mutation=True)
def epi_sankey(sankey_data,disease_or_gard_id):
    gathered, relevant, epidemiologic = sankey_data
    
    fig = go.Figure(data=[go.Sankey(
        node = dict(
          pad = 15,
          thickness = 20,
          line = dict(color = "white", width = 0.5),
          label = ["PubMed IDs Gathered", "Irrelevant Abstracts","Relevant Abstracts Gathered","Epidemiologic Abstracts","Not Epidemiologic"],
          color = "purple"
        ),
        #label = ["A1", "A2", "B1", "B2", "C1", "C2"]
        link = dict(
          source = [0, 0, 2, 2],
          target = [2, 1, 3, 4],
          value = [relevant, gathered-relevant, epidemiologic, relevant-epidemiologic]
      ))])
    fig.update_layout(
    hovermode = 'x',
    title="Search for the Epidemiology of "+disease_or_gard_id,
    font=dict(size = 10, color = 'black'),
)
    
    return fig

#### BEGIN APP ####

with st.spinner('Loading Epidemiology Models and Dependencies...'):
    classify_model_vars, NER_pipeline, entity_classes, GARD_dict, max_length = load_models_experimental()
    #classify_tokenizer, classify_model, NER_pipeline, entity_classes, GARD_dict, max_length = load_models()
    #Load spaCy models which cannot be cached due to hash function error
    #nlp = spacy.load('en_core_web_lg')
    #nlpSci = spacy.load("en_ner_bc5cdr_md")
    #nlpSci2 = spacy.load('en_ner_bionlp13cg_md')
    #classify_model_vars = (nlp, nlpSci, nlpSci2, classify_model, classify_tokenizer)

loaded = st.success('All Models and Dependencies Loaded!')

st.markdown("Input a rare disease term or GARD ID.")

disease_or_gard_id = st.text_input('')

loaded.empty()

st.markdown("Examples of rare diseases include [**Fellman syndrome**](https://rarediseases.info.nih.gov/diseases/1/gracile-syndrome), [**Classic Homocystinuria**](https://rarediseases.info.nih.gov/diseases/6667/classic-homocystinuria), [**7383**](https://rarediseases.info.nih.gov/diseases/7383/phenylketonuria), and [**GARD:0009941**](https://rarediseases.info.nih.gov/diseases/9941/fshmd1a).")

st.markdown("A full list of rare diseases tracked by GARD can be found [here](https://rarediseases.info.nih.gov/diseases/browse-by-first-letter).")

if disease_or_gard_id:
    df, sankey_data = extract_abs.streamlit_extraction(disease_or_gard_id, max_results, filtering,
                                NER_pipeline, entity_classes, 
                                extract_diseases,GARD_dict, max_length, 
                                classify_model_vars)
    st.dataframe(df, height=100)
    csv = convert_df(df)
    st.download_button(
        label="Download epidemiology results for "+disease_or_gard_id+" as CSV",
        data = csv,
        file_name=disease_or_gard_id+'.csv',
        mime='text/csv',
        )
    #st.dataframe(data=None, width=None, height=None)
    fig = epi_sankey(sankey_data,disease_or_gard_id)
        
    #if st.button('Display Sankey Diagram of Automated Search'):
    st.plotly_chart(fig, use_container_width=True)
# st.code(body, language="python")