streamlit_app.py

import streamlit as st
import os
import pickle
import pandas as pd
from PIL import Image
from rdkit import Chem
from rdkit.Chem import AllChem, Draw
from padelpy import padeldescriptor

# Page configuration
st.set_page_config(
  page_title='PARP1pred',
  page_icon='💊',
  initial_sidebar_state='expanded')

# Session state
if 'smiles_input' not in st.session_state:
  st.session_state.smiles_input = ''

# Utilities
if os.path.isfile('molecule.smi'):
  os.remove('molecule.smi') 
  

# The App    
st.title('💊 PARP1pred app')
st.info('PARP1pred allow users to predict whether a query molecule is active/inactive towards the PARP1 target protein.')

tab1,tab2,tab3,tab4,tab5,tab6,tab7 = st.tabs(['Main', 'About', 'What is PARP1?', 'Dataset', 'Model performance', 'Python libraries', 'Citing us'])


with tab1:
  if st.session_state.smiles_input == '':
    
    with st.form('my_form'):
      st.subheader('Predict PARP1 inhibitory activity')

      smiles_txt = st.text_input('Enter SMILES notation', st.session_state.smiles_input)
      st.session_state.smiles_input = smiles_txt

      with st.expander('Example SMILES'):
        st.code('O=C(c1cc(Cc2n[nH]c(=O)c3ccccc23)ccc1F)N1CCN(C(=O)C2CC2)CC1')
      
      submit_button = st.form_submit_button('Submit')

      
      
      if submit_button:
        st.subheader('⚛️ Input molecule:')
        with st.expander('Show SMILES', expanded=True):
          #st.write('**SMILES**')
          st.text(st.session_state.smiles_input)

        with st.expander('Show chemical structures', expanded=True):
          #st.write('**Chemical structure**')
          smi = Chem.MolFromSmiles(st.session_state.smiles_input)
          Chem.Draw.MolToFile(smi, 'molecule.png', width=900)
          mol_image = Image.open('molecule.png')
          st.image(mol_image)

      # Input SMILES saved to file
      f = open('molecule.smi', 'w')
      f.write(f'{st.session_state.smiles_input}\tmol_001')
      f.close()


      # Compute PADEL descriptors
      if st.session_state.smiles_input != '':
        st.subheader('🔢 Descriptors')
        if os.path.isfile('molecule.smi'):
          padeldescriptor(mol_dir='molecule.smi', 
                            d_file='descriptors.csv',
                            descriptortypes='data/PubchemFingerprinter.xml', 
                            detectaromaticity=True,
                            standardizenitro=True,
                            standardizetautomers=True,
                            threads=2,
                            removesalt=True,
                            log=True,
                            fingerprints=True)

        descriptors = pd.read_csv('descriptors.csv')
        descriptors.drop('Name', axis=1, inplace=True)

        with st.expander('Show full set of descriptors as calculated for query molecule'):
          #st.write('**Full set of descriptors (calculated for query molecule)**')
          st.write(descriptors)
          st.write(descriptors.shape)


      # Load descriptor subset used in trained model
      if st.session_state.smiles_input != '':
        model = pickle.load(open('data/oversampling_PubChem_RandomForestClassifier.pkl', 'rb'))
        pubchem_subset = model.feature_names_in_

        query_desc_1 = descriptors.columns.difference(pubchem_subset)
        query_desc_2 = descriptors.drop(query_desc_1, axis=1)

        with st.expander('Show subset of descriptors as used in trained model'):
          #st.write('**Subset of descriptors (used in trained model)**')
          st.write(query_desc_2)
          st.write(query_desc_2.shape)


      # Read in saved classification model
      if st.session_state.smiles_input != '':
        st.subheader('🤖 Predictions')
        pred = int(model.predict(query_desc_2))
        if pred == 0:
          st.error('Inactive')
        if pred == 1:
          st.success('Active')

        
with tab2:
  coverimage = Image.open('PARP1pred.jpg')
  st.image(coverimage)
with tab3:
  st.header('What is PARP1?')
  st.write('Poly (ADP-ribose) polymerase-1 (PARP-1) is an enzyme that catalyzes the ADP-ribosylation of a specific protein and plays a vital role in DNA repair. It has become an attractive target as inhibition of PARP-1 causes a toxic accumulation of DNA double strand breaks in cancer cells, particularly those with BRCA1/2 deficiency, which are found in breast, ovarian, prostate, and pancreatic cancers.')
with tab4:
  st.header('Dataset')
  st.write('''
    In our work, we retrieved a human PARP-1 biological dataset from the ChEMBL database. The data was curated and resulted in a non-redundant set of 2,018 PARP-1 inhibitors, which can be divided into:
    - 1,720 active compounds
    - 298 inactive compounds
    ''')
with tab5:
  st.header('Model performance')
  st.write('We selected PubChem as a molecular fingerprint and used a random forest with an oversampling approach to construct the best model. The Matthews correlation coefficients for training, cross-validation, and test sets were 1.00, 0.96, and 0.74, respectively.')
with tab6:
  st.header('Python libraries')
  st.markdown('''
    This app is based on the following Python libraries:
    - `streamlit`
    - `pandas`
    - `rdkit`
    - `padelpy`
  ''')
with tab7:
  st.markdown('T. Lerksuthirat, S. Chitphuk, W. Stitchantrakul, D. Dejsuphong, A.A. Malik, C. Nantasenamat, PARP1PRED: A web server for screening the bioactivity of inhibitors against DNA repair enzyme PARP-1, ***EXCLI Journal*** (2023) DOI: https://doi.org/10.17179/excli2022-5602.')