Spaces:

dmsaylor
/

CHRIS

Running

App Files Files Community
CHRIS / ChemID.py
3v324v23
Claude Code
Remove dead code and untrack DS_Store
35f6c3d 11 days ago
Raw History Blame Contribute Delete
56.7 kB
import sys,os,glob,re,string
from collections import Counter
from operator import itemgetter
import nltk
import pandas as pd
import numpy as np
import pubchempy as pcp
import cirpy
import chemicals

import bs4
import urllib
import requests
import json
import logging
import time
import threading
import urllib.request
from xml.etree import ElementTree

_CIRPY_CHECK_TIMEOUT = float(os.environ.get('CIRPY_CHECK_TIMEOUT', '5'))
_CIRPY_RECHECK_INTERVAL = 24 * 60 * 60
_logger = logging.getLogger(__name__)

def _check_cirpy_access(reason="on_demand"):
    """Probe CACTUS with a timeout, without changing CIRpy's shared urlopen."""
    started = time.monotonic()
    context = (time.strftime('%Y-%m-%dT%H:%M:%SZ', time.gmtime()), os.getpid(), reason)
    try:
        url = cirpy.construct_api_url('water', 'smiles')
        # Log actual probes (not cached uses), including successful forced tests.
        _logger.warning(
            'CIRPY_CHECK start_utc=%s pid=%s reason=%s event=start url=%s timeout_s=%s',
            *context, url, _CIRPY_CHECK_TIMEOUT,
        )
        with urllib.request.urlopen(url, timeout=_CIRPY_CHECK_TIMEOUT) as response:
            root = ElementTree.parse(response).getroot()
        if not any((item.text or '').strip() for item in root.findall('.//data/item')):
            raise ValueError('CACTUS returned no SMILES result')
    except Exception as exc:
        _logger.warning(
            'CIRPY_CHECK start_utc=%s pid=%s reason=%s event=end available=False '
            'elapsed_s=%.3f error=%s: %s',
            *context, time.monotonic() - started, type(exc).__name__, exc,
        )
        return False

    _logger.warning(
        'CIRPY_CHECK start_utc=%s pid=%s reason=%s event=end available=True elapsed_s=%.3f',
        *context, time.monotonic() - started,
    )
    return True

_force_disable_cirpy = os.environ.get(
    "DISABLE_CIRPY", ""
).strip().lower() in {"1", "true", "yes"}

# Per-process cache: importing ChemID does not contact CACTUS.
try_cirpy = False
_cirpy_last_check = None
_cirpy_check_lock = threading.Lock()

def _cirpy_available(force=False):
    """Check on first use or after 24 hours; cache failures as well as successes."""
    global try_cirpy, _cirpy_last_check
    if _force_disable_cirpy:
        return False
    # Prevent duplicate ordinary probes from concurrent requests in this process.
    with _cirpy_check_lock:
        if (force or _cirpy_last_check is None
                or time.monotonic() - _cirpy_last_check >= _CIRPY_RECHECK_INTERVAL):
            reason = 'diagnostic_identifier' if force else 'on_demand'
            try_cirpy = _check_cirpy_access(reason=reason)
            _cirpy_last_check = time.monotonic()
        return try_cirpy

ORGANIC_ATOM_SET = {5, 6, 7, 8, 9, 15, 16, 17, 35, 53}
METAL_ATOM_SET = set([3,4,11,12,13] + list(range(19,31+1)) + list(range(37,50+1)) + list(range(55,84+1)) + list(range(87,114+1)) + [116])
with open('data/ceramics_list.txt', 'r') as fp:
    lines = fp.readlines()
CERAMICS_SET = {line.strip() for line in lines}
with open('data/salt_list.txt', 'r') as fp:
    lines = fp.readlines()
SALT_SET = {line.strip() for line in lines}
CERAMIC_MP_CUTOFF_C = 700.0

ERROR_CODES = {0:None, 1:'Structure could not be determined from the identifier', 2:'Invalid SMILES code', 3:'Invalid CAS number', 4:'Invalid identifier type selected', 5:'Invalid InChI'}

## not sure if this will be possible on pythonanywhere; use this flag to disable related code blocks
try_dsstox = False
if try_dsstox:
    import selenium
    import selenium.webdriver
    from selenium.webdriver.chrome.options import Options
    from selenium.webdriver.chrome.service import Service
    ## NOTE this does not seem very robust
    uname = os.uname()
    if uname.sysname == 'Darwin':
        # mac
        import chromedriver_binary
        driver_exe = chromedriver_binary.chromedriver_filename
    elif uname.sysname == 'Linux':
        # pythonanywhere
        driver_exe = '/usr/local/bin/chromedriver'
    else:
        # fallback...
        import chromedriver_binary
        driver_exe = chromedriver_binary.chromedriver_filename

from io import BytesIO
from PIL import ImageOps
import base64

import rdkit
from rdkit.Chem import AllChem as Chem
from rdkit.Chem import Descriptors,Draw,Crippen
from ceramic_classifier import predict_high_mp

## add custom chemical definitions (i.e., to correct confusion between methane and carbon)
db = chemicals.identifiers.get_pubchem_db()
db.load('data/custom_chemicals_db.tsv')
## load experimental and predicted properties
#dfmp_expt = pd.read_excel('PHYSPROP_MP_data.xlsx')
dfmp_expt = pd.read_csv('data/PHYSPROP_MP_data.tsv', sep='\t')
#dfmp_pred = pd.read_excel('DSSTOX_MP_pred_data.xlsx')
#df_pred = pd.read_excel('Comptox_pred_data.xlsx')
df_pred = pd.read_csv('data/Comptox_pred_data.tsv', sep='\t')

## OPERA melting point model
import dill as pickle
import sklearn
import sklearn.neighbors
import sklearn.metrics
import padelpy
from functions import weight_func
class opera_data_mp():
    n_neighbors = 5
    #weight_factor = 5e-3
    desc_list = ['SHBd', 'nN', 'maxHBd', 'ATSC1v', 'AATS1i', 'TopoPSA', 'nT6Ring', 'nHBDon', 'WTPT-5', 'minHBd', 'nHBint2', 'IC0', 'MLFER_S', 'MLFER_BO', 'WTPT-3']
    with open('mp/model-opera-knn.pkl', 'rb') as fp:
        knn_all = pickle.load(fp)
    knn_all.weights = weight_func # fix weird problem on pythonanywhere...
    with open('mp/model-opera-scalerX.pkl', 'rb') as fp:
        scaler_X = pickle.load(fp)
    with open('mp/model-opera-scalerY.pkl', 'rb') as fp:
        scaler_y = pickle.load(fp)
my_opera_data_mp = opera_data_mp()

def ResolveChemical(chemName, IDtype, debug=False, get_properties=['logp','rho','mp']):

    #LogP_func = Crippen.MolLogP
    LogP_func = getLogP
    
    # remove excess whitespace
    chemName = chemName.strip()

    # Diagnostic input: refresh availability even if this name resolves locally.
    if IDtype == 'common' and ' '.join(chemName.split()).casefold() == 'irganox 3114':
        _cirpy_available(force=True)
    
    name = None
    smiles = None
    cas = None
    Mw = None
    LogP = None
    rho = None
    mp = None
    im64 = None
    mp_origin = None
    rho_origin = None
    LogP_origin = None
    error = 0

    if IDtype == 'CAS':
        cas = chemName

        if not is_cas(cas):
            error = 3 #invalid cas
            return (name, cas, smiles, Mw, LogP, rho, mp, im64, error)

        smiles = cas2smiles(cas)

        if smiles:
            name = cas2name(cas)
            if 'rho' in get_properties:
                rho, rho_origin = string2density(cas)
            if 'mp' in get_properties:
                mp, mp_origin = mol2mp(cas, name, smiles)
            if 'rho' in get_properties and pd.isna(rho) and name:
                rho, rho_origin = string2density(name)
            try:
                mol = Chem.MolFromSmiles(smiles)
            except:
                error = 2 #invalid smiles
            if mol:
                Mw = Descriptors.MolWt(mol)
                if 'logp' in get_properties:
                    LogP, LogP_origin = LogP_func(cas, mol)
                im = ImageFromSmiles(smiles)
                im64 = Imageto64(im)
            else:
                error = 2 #invalid smiles
        else:
            error = 1 # no smiles found
    elif IDtype == 'SMILES':
        smiles = chemName

        try:
            mol = Chem.MolFromSmiles(smiles)
        except:
            error = 2
        if mol:
            Mw = Descriptors.MolWt(mol)
            if 'logp' in get_properties:
                LogP, LogP_origin = LogP_func(cas, mol)
            im = ImageFromSmiles(smiles)
            im64 = Imageto64(im)
        else:
            error = 2

        # if SMILES is not valid, skip the other stuff
        if not error:
            name = smiles2name(smiles)
            if name:
                cas = name2cas(name)
                if 'rho' in get_properties:
                    rho, rho_origin = string2density(name)
            if 'rho' in get_properties and pd.isna(rho) and cas:
                rho, rho_origin = string2density(cas)

            if 'mp' in get_properties:
                mp, mp_origin = mol2mp(cas, name, smiles)
    elif IDtype == 'InChI':
        inchi = chemName

        try:
            mol = Chem.MolFromInchi(inchi)
        except:
            mol = None

        if mol:
            smiles = Chem.MolToSmiles(mol, isomericSmiles=True)
            Mw = Descriptors.MolWt(mol)
            if 'logp' in get_properties:
                LogP, LogP_origin = LogP_func(cas, mol)
            im = ImageFromSmiles(smiles)
            im64 = Imageto64(im)

            name = smiles2name(smiles)
            if name:
                cas = name2cas(name)
                if 'rho' in get_properties:
                    rho, rho_origin = string2density(name)
            if 'rho' in get_properties and pd.isna(rho) and cas:
                rho, rho_origin = string2density(cas)

            if 'mp' in get_properties:
                mp, mp_origin = mol2mp(cas, name, smiles)
        else:
            error = 5
    elif IDtype == 'common':
        name = chemName

        name, name_origin = name2iupac(name)
        smiles = name2smiles(name)
        cas = name2cas(name)
        if not smiles:
            smiles = cas2smiles(cas)

        if 'rho' in get_properties and pd.isna(rho) and cas:
            rho, rho_origin = string2density(cas)
        if 'rho' in get_properties and pd.isna(rho):
            rho, rho_origin = string2density(name)
        if 'rho' in get_properties and pd.isna(rho):
            # try this because sometimes iupac names don't work
            rho, rho_origin = string2density(chemName)

        if smiles:
            if 'mp' in get_properties:
                mp, mp_origin = mol2mp(cas, name, smiles)
            try:
                mol = Chem.MolFromSmiles(smiles)
            except:
                error = 2
            if mol:
                Mw = Descriptors.MolWt(mol)
                if 'logp' in get_properties:
                    LogP, LogP_origin = LogP_func(cas, mol)
                im = ImageFromSmiles(smiles)
                im64 = Imageto64(im)
            else:
                error = 2
        else:
            error = 1
    else:
        ## should never be here
        name = None
        smiles = None
        cas = None
        Mw = None
        LogP = None
        rho = None
        mp = None
        im64 = None
        error = 4 # invalid IDtype selection, probably not possible

    # if we couldn't find a name or CAS (but do have SMILES)
    if not error:
        if not name:
            name = 'Not found'
        if not cas:
            cas = 'Not found'
    
    if mp is not None:
        mp = float(mp)

    if debug:
        return (name, cas, smiles, Mw, LogP, LogP_origin, rho, rho_origin, mp, mp_origin, im64, error)
    else:
        return (name, cas, smiles, Mw, LogP, rho, mp, im64, error)

def _is_elemental_carbon_mol(mol):
    # RDKit distinguishes elemental carbon, [C], from methane, C, by H count.
    if mol is None or mol.GetNumAtoms() != 1:
        return False
    atom = mol.GetAtomWithIdx(0)
    return atom.GetAtomicNum() == 6 and atom.GetTotalNumHs() == 0

def _material_precheck(smiles):
    # Classify anything that does not require a melting-point lookup.
    is_ceramic = False
    mol = Chem.MolFromSmiles(smiles)
    atom_num_list = [a.GetAtomicNum() for a in mol.GetAtoms()]
    is_metal = set(atom_num_list) <= METAL_ATOM_SET
    needs_mp = False
    if not is_metal:
        # check composition against list of ceramics/salts
        elements = ','.join(sorted(set([a.GetSymbol() for a in mol.GetAtoms()])))
        if elements in CERAMICS_SET:
            is_ceramic = True
        if elements in SALT_SET:
            is_ceramic = True
        if not is_ceramic:
            # Ordinary carbon-containing molecules are non-ceramic. Elemental
            # carbon ([C], e.g. graphite/diamond) is an explicit exception so
            # its experimental/tabulated MP can resolve the classification.
            has_carbon = 6 in atom_num_list
            is_elemental_carbon = _is_elemental_carbon_mol(mol)
            needs_mp = (not has_carbon) or is_elemental_carbon
    return is_metal, is_ceramic, needs_mp

def CeramicOrMetal(smiles,mp):
    # Backwards-compatible material classification when MP is already known.
    is_metal, is_ceramic, needs_mp = _material_precheck(smiles)
    if needs_mp and (mp is not None) and mp > CERAMIC_MP_CUTOFF_C:
        is_ceramic = True
    return is_metal, is_ceramic

def ClassifyMaterial(cas, name, smiles):
    # Avoid MP/model lookups unless they can change the ceramic classification.
    is_metal, is_ceramic, needs_mp = _material_precheck(smiles)
    mp, mp_origin = None, None
    mp_searched = False
    if not needs_mp:
        return is_metal, is_ceramic, mp, mp_origin, mp_searched

    mp_searched = True
    if cas == 'Not found':
        cas = None
    if name == 'Not found':
        name = None

    # 1. Prefer a local experimental/tabulated MP when available.
    if cas:
        try:
            mp, mp_origin = string2mp(
                cas, local_only=True, include_predictions=False
            )
        except Exception:
            mp, mp_origin = None, None
    if mp is not None:
        is_ceramic = mp > CERAMIC_MP_CUTOFF_C
        return is_metal, is_ceramic, mp, mp_origin, mp_searched

    # 2. Use the local Magpie classifier for the validated carbon-free domain.
    # Elemental carbon is deliberately excluded from the classifier and is
    # resolved only from experimental/tabulated MP sources.
    mol = Chem.MolFromSmiles(smiles)
    is_elemental_carbon = _is_elemental_carbon_mol(mol)
    prediction = None
    if not is_elemental_carbon:
        try:
            prediction = predict_high_mp(cas=cas, name=name, smiles=smiles)
        except Exception:
            prediction = None
    if prediction is not None:
        is_ceramic = prediction.is_high_mp
        # This is classifier provenance, not a physical MP prediction.
        mp_origin = prediction.origin
        return is_metal, is_ceramic, None, mp_origin, mp_searched

    # 3. If the classifier cannot make a safe prediction, try remote
    # experimental/tabulated MP sources.  Do not use predictive MP sources in
    # the ceramic fallback.
    if cas:
        try:
            mp, mp_origin = string2mp(cas, include_predictions=True)
        except Exception:
            mp, mp_origin = None, None
    if mp is None and name:
        try:
            mp, mp_origin = string2mp(name, include_predictions=True)
        except Exception:
            mp, mp_origin = None, None
    if mp is not None:
        is_ceramic = mp > CERAMIC_MP_CUTOFF_C

    # If every source is unavailable, the conservative fallback is non-ceramic.
    return is_metal, is_ceramic, mp, mp_origin, mp_searched

#Generates an image of the molecule represented by the SMILES code given.
#Returns None if the image cannot be generated. From https://github.com/ronaldo-prata/flask-test/blob/master/functions.py

def ImageFromSmiles(smiles):
    image = None
    if type(smiles) is str:
        try:
            if smiles == 'C1=CC=C2C(=C1)C3=NC4=NC(=NC5=C6C=CC=CC6=C([N-]5)N=C7C8=CC=CC=C8C(=N7)N=C2[N-]3)C9=CC=CC=C94.[Mn+2]':
                mol = next(Chem.SDMolSupplier('data/MnPC.sdf', removeHs=False))
                image = Draw.MolToImage(mol, size=(350, 350))
            else:
                image = Draw.MolToImage(Chem.MolFromSmiles(smiles), size=(350, 350))
        except ValueError:
            pass
    return image

#Trims the image into a box, removing any excess white background.
#The box cannot be smaller than 400x400. This is done due to the difference in quality in the images generated by MolToImage,
#  if the size is too small (300x300), big molecules are too low quality, but if the size is too big (1000*1000), small molecules appear zoomed out.

def wTrim(img):
    bbox = ImageOps.invert(img).getbbox()
    crop = (bbox[0], bbox[1], bbox[2], bbox[3])

    return img.crop(crop)

#Converts a PIL image into its base64 representation.
def Imageto64(img):
    img = wTrim(img)
    buf = BytesIO()
    img.save(buf, format="PNG")
    pngImageB64String = "data:image/png;base64,"
    pngImageB64String += base64.b64encode(buf.getvalue()).decode("utf-8")

    return pngImageB64String

# function to convert SMILES to name
def smiles2name(smiles):
    name = None
    # first try chemicals package
    try:
        cm = chemicals.search_chemical(f'SMILES={smiles}')
        if cm.iupac_name:
            name = cm.iupac_name
        elif cm.common_name:
            name = cm.common_name
    except:
        name = None
    # then try pubchem for compounds
    if not name:
        try:
            compounds = pcp.get_compounds(smiles, namespace='smiles')
            c = compounds[0]
            name = c.iupac_name
            if not name:
                # have seen empty iupac_name before, try synonyms if this happens
                name = c.synonyms[0]
        except:
            name = None
    # next try cirpy
    if not name and _cirpy_available():
        try:
            name = cirpy.resolve(smiles, 'iupac_name')
        except:
            name = None
    if type(name) is list:
        name = name[0]
    # finally try it as a pubchem substance
    if not name:
        try:
            compounds = pcp.get_substances(smiles, namespace='smiles')
            # sometimes there are multiple substances, and multiple synonyms per substance
            allsyns = [syn for c in compounds for syn in c.iupac_name if smiles not in syn]
            # choose the most common synonym
            fd = nltk.FreqDist(allsyns)
            name = fd.most_common(1)[0][0]
        except:
            name = None
    return name

# function to convert CAS to SMILES
def cas2smiles(cas):
    smiles = None
    # first try chemicals package
    try:
        cm = chemicals.search_chemical(cas)
        smiles = cm.smiles
    except:
        smiles = None
    # then try pubchem for compounds
    if not smiles:
        try:
            compounds = pcp.get_compounds(cas, namespace='name')
            c = compounds[0]
            smiles = c.isomeric_smiles
        except:
            smiles = None
    # next try cirpy
    if not smiles and _cirpy_available():
        try:
            smiles = cirpy.resolve(cas, 'smiles')
        except:
            smiles = None
    if type(smiles) is list:
        smiles = smiles[0]
    # finally try it as a pubchem substance
    if not smiles:
        try:
            compounds = pcp.get_substances(cas, namespace='name')
            # sometimes there are multiple substances, and multiple synonyms per substance
            allsyns = [syn for c in compounds for syn in c.isomeric_smiles if cas not in syn]
            # choose the most common synonym
            fd = nltk.FreqDist(allsyns)
            smiles = fd.most_common(1)[0][0]
        except:
            smiles = None
    return smiles

# function to convert cas to name
def cas2name(cas):
    name = None
    #if not is_cas(cas):
    #    name = 'INVALID CAS'
    # first try chemicals package
    try:
        cm = chemicals.search_chemical(cas)
        if cm.iupac_name:
            name = cm.iupac_name
        elif cm.common_name:
            name = cm.common_name
    except:
        name = None
    # then try cirpy
    if not name and _cirpy_available():
        try:
            name = cirpy.resolve(cas, 'iupac_name')
        except:
            name = None
    if type(name) is list:
        name = name[0]
    # next try pubchem for compounds
    if not name:
        try:
            compounds = pcp.get_compounds(cas, namespace='name')
            c = compounds[0]
            name = c.iupac_name
            if not name:
                # have seen empty iupac_name before, try synonyms if this happens
                name = c.synonyms[0]
        except:
            name = None
    return name

# function to convert chemical name to iupac name
def name2iupac(string):
    name = None
    origin = None
    # try chemicals package
    try:
        cm = chemicals.search_chemical(string)
        if cm.iupac_name:
            name = cm.iupac_name
        elif cm.common_name:
            name = cm.common_name
        origin = 'chemicals'
    except:
        name = None
        origin = None
    # try pubchem for compounds
    if not name:
        try:
            compounds = pcp.get_compounds(string, namespace='name')
            c = compounds[0]
            name = c.iupac_name
            if not name:
                # have seen empty iupac_name before, try synonyms if this happens
                name = c.synonyms[0]
            origin = 'PubChem'
        except:
            name = None
            origin = None
    # next try cirpy
    if not name and _cirpy_available():
        try:
            #name = cirpy.resolve(string, 'names')
            name = cirpy.resolve(string, 'iupac_name')
            if name: origin = 'CIRPY'
        except:
            name = None
            origin = None
    if type(name) is list:
        name = name[0]
    # now try it as a pubchem substance
    if not name:
        try:
            compounds = pcp.get_substances(string, namespace='name')
            # sometimes there are multiple substances, and multiple synonyms per substance
            allsyns = [syn for c in compounds for syn in c.synonyms if string not in syn]
            # choose the most common synonym
            fd = nltk.FreqDist(allsyns)
            name = fd.most_common(1)[0][0]
            origin = 'PubChem/substance'
        except:
            name = None
            origin = None
    # strip all spaces and try again...
    if not name:
        string_strip = re.sub(' ','',string)
        # first try pubchem for compounds
        try:
            compounds = pcp.get_compounds(string_strip, namespace='name')
            c = compounds[0]
            name = c.iupac_name
            if not name:
                # have seen empty iupac_name before, try synonyms if this happens
                name = c.synonyms[0]
            origin = 'PubChem'
        except:
            name = None
            origin = None
        # next try cirpy
        if not name and _cirpy_available():
            try:
                #name = cirpy.resolve(string_strip, 'names')
                name = cirpy.resolve(string, 'iupac_name')
                if name: origin = 'CIRPY'
            except:
                name = None
                origin = None
        if type(name) is list:
            name = name[0]
    return name, origin

# function to convert name to cas
def name2cas(name):
    cas = None
    # try chemicals package
    try:
        cm = chemicals.search_chemical(name)
        cas = cm.CASs
    except:
        cas = None
    # then try cirpy
    if not cas and _cirpy_available():
        try:
            cas = cirpy.resolve(name, 'cas')
        except:
            cas = None
    if type(cas) is list:
        cas.sort(key=lambda s: np.array(s.split('-'), dtype=int).sum())
        cas = cas[0]
    # next try pubchem for compounds
    if not cas:
        try:
            compounds = pcp.get_compounds(name, namespace='name')
            c = compounds[0]
            syns = c.synonyms
            possible_cas = [syn for syn in syns if is_cas(syn)]
            # if multiple choose option with smallest sum of digits
            possible_cas.sort(key=lambda s: np.array(s.split('-'), dtype=int).sum())
            cas = possible_cas[0]
        except:
            cas = None
    return cas

# function to convert name to SMILES
def name2smiles(name):
    smiles = None
    # first try chemicals package
    try:
        cm = chemicals.search_chemical(name)
        smiles = cm.smiles
    except:
        smiles = None
    # then try pubchem for compounds
    if not smiles:
        try:
            compounds = pcp.get_compounds(name, namespace='name')
            c = compounds[0]
            smiles = c.isomeric_smiles
        except:
            smiles = None
    # next try cirpy
    if not smiles and _cirpy_available():
        try:
            smiles = cirpy.resolve(name, 'smiles')
        except:
            smiles = None
    if type(smiles) is list:
        smiles = smiles[0]
    # then try it as a pubchem substance
    if not smiles:
        try:
            compounds = pcp.get_substances(name, namespace='name')
            # sometimes there are multiple substances, and multiple synonyms per substance
            allsyns = [syn for c in compounds for syn in c.isomeric_smiles if name not in syn]
            # choose the most common synonym
            fd = nltk.FreqDist(allsyns)
            smiles = fd.most_common(1)[0][0]
        except:
            smiles = None
    # finally try to resolve SMILES from name using OPSIN
    #if not smiles:
    #    try:
    #        with open('opsin.tmp.1', 'w') as fp:
    #            fp.write(name)
    #        os.system('java -jar /Users/robert.elder/software/utils/opsin-2.4.0-jar-with-dependencies.jar -osmi opsin.tmp.1 opsin.tmp.2 &> /dev/null')
    #        with open('opsin.tmp.2') as fp:
    #            smiles = fp.read()
    #        if smiles == '\n':
    #            smiles = None
    #        smiles = smiles.strip() #remove trailing newline
    #    except KeyboardInterrupt:
    #        raise
    #    except:
    #        smiles = None
    return smiles

def check_cas(cas):
    n1,n2,n3 = cas.split('-')
    # combine and flip first 2 numbers
    tmp = ''.join([n1,n2])[::-1]
    # sum of number*position in string
    check = sum([i*int(tmp[i-1]) for i in range(1,len(tmp)+1)])
    # mod 10
    check = check%10
    # if these match, then it's a legit cas number
    return check == int(n3)

def is_cas(cas):
    try:
        return check_cas(cas)
    except:
        return False

def string2density(name):
    rho, rho_origin = None, None
    # predicted values from TEST (CompTox dashboard)
    if is_cas(name):
        mask = df_pred['CASRN'] == name
        if sum(mask):
            rho = float(df_pred[mask]['DENSITY_G/CM^3_TEST_PRED'].iloc[0])
            rho_origin = 'comptox/pred'
    # try to scrape from PubChem
    if pd.isna(rho):
        content = None
        try:
            compounds = pcp.get_compounds(name, namespace='name')
            c = compounds[0]
            cid = c.cid
            url = f'https://pubchem.ncbi.nlm.nih.gov/rest/pug_view/data/compound/{cid}/JSON'
            #fid = urllib.request.urlopen(url)
            #webpage = fid.read().decode('utf-8')
            webpage = requests.get(url).text
            content = json.loads(webpage)
        except:
            pass
        if content:
            rho_list = []
            for i in content['Record']['Section']:
                if i['TOCHeading'] == 'Chemical and Physical Properties':
                    for j in i['Section']:
                        if j['TOCHeading'] == 'Experimental Properties':
                            for k in j['Section']:
                                if k['TOCHeading'] == 'Density':
                                    for ii in k['Information']:
                                        try:
                                            rho_string = ii['Value']['StringWithMarkup'][0]['String']
                                            rho_string = rho_string.replace('Relative density (water = 1): ', '')
                                            #print(rho_string)
                                            #tmp_rho = re.match('(?:\d+(?:\.\d*)?|\.\d+)',rho_string)
                                            m = re.match(r'((?:\d+(?:\.\d*)?|\.\d+))(?:-((?:\d+(?:\.\d*)?|\.\d+)))?',rho_string)
                                            groups = m.groups()
                                            if len(groups):
                                                for g in groups:
                                                    try:
                                                        tmp_rho = float(g)
                                                        rho_list.append(tmp_rho)
                                                    except:
                                                        continue
                                        except:
                                            continue
            if rho_list:
                ## remove outliers using interquartile range (IQR)
                rho_list = np.array(rho_list)
                q75,q25 = np.percentile(rho_list,[75,25])
                intr_qr = q75-q25
                hi = q75+(1.5*intr_qr)
                lo = q25-(1.5*intr_qr)
                mask = (rho_list <= hi) & (rho_list >= lo)
                rho_list = rho_list[mask]
                rho = np.mean(rho_list)
                rho_origin = 'pubchem'
            else:
                rho, rho_origin = None, None
    # try to scrape from DSSTOX
    if try_dsstox:
        if pd.isna(rho):
            dtxsid = None
            try:
                # try to find it via the dsstox dashboard
                name_urlsafe = urllib.parse.quote(name)
                url = f'https://comptox.epa.gov/dashboard/search-results?input_type=synonym_substring&inputs={name_urlsafe}'
                fid = urllib.request.urlopen(url)
                webpage = fid.read().decode('utf-8')
                hits = re.findall(r'DTXSID[0-9]+', webpage)
                if len(hits):
                    dtxsid = hits[0]
            except:
                pass
            if dtxsid:
                mysoup = None
                url = f'https://comptox.epa.gov/dashboard/chemical/properties/{dtxsid}'
                #print(url)
                try:
                    options = Options()
                    options.add_argument("--headless") # runs in background instead of showing browser window
                    service = Service(driver_exe)
                    driver = selenium.webdriver.Chrome(service=service, options=options)
                    driver.set_page_load_timeout(15)
                    driver.get(url)
                    webpage = driver.page_source
                    driver.quit()
                    mysoup = bs4.BeautifulSoup(webpage, features='lxml')
                except KeyboardInterrupt:
                    raise
                except:
                    pass
                if mysoup:
                    ifound = None
                    # column of property names
                    rows = mysoup.find_all('div', attrs={'col-id':'property'})
                    for i,row in enumerate(rows):
                        if 'Density' in row.text:
                            ifound = i
                            break
                    if ifound:
                        rows = mysoup.find_all('div', attrs={'col-id':'exavg'})
                        text = rows[ifound].text
                        value = re.sub(r' \([0-9]*\)', '', text.strip())
                        try:
                            rho = float(value)
                            rho_origin = 'dsstox/expt'
                        except:
                            rho, rho_origin = None, None
                        if pd.isna(rho):
                            rows = mysoup.find_all('div', attrs={'col-id':'predavg'})
                            text = rows[ifound].text
                            value = re.sub(r' \([0-9]*\)', '', text.strip())
                            try:
                                rho = float(value)
                                rho_origin = 'dsstox/pred'
                            except:
                                rho, rho_origin = None, None
                    else:
                        rho, rho_origin = None, None
            else:
                rho, rho_origin = None, None
    if pd.isna(rho): rho,rho_origin = None, None
    return rho, rho_origin

def return_non_duplicate_index(tuples): ##Given a list of sets return index of non_duplicate items
    ## from https://github.com/curieshicy/JRgui/
    ##step 1, create a new tuple, named "new_tuples"
    new_tuples = [] ##the elements are the sets
    for i in tuples:
        for j in i:
            new_tuples.append(set(j))
    ##step 2, create a dictionary storing one to one relationship between new_tuple and old_tuple
    values = []
    for index, item in enumerate(tuples):
        if len(item) == 1:
            values.append(index)
        else:
            for i in [index]*len(item):
                values.append(i)
    keys = [i for i in range(len(new_tuples))]
    dict_tuples = {}  ## {0:0, 1:1, 2:2, 3:3, 4:3, 5:3, 6:4, 7:4, 8:4, 9:5, 10:6, 11:7, 12:8}
    for i, j in zip(keys, values):
        dict_tuples[i] = j
    ##step 3, remove duplicates in sets terminology    
    remove_index = []
    for index_1, item in enumerate(new_tuples): ##starting from beginning
        for index_2 in range(index_1 + 1, len(new_tuples)): ##loop over the rest items
            if len(item & new_tuples[index_2]) != 0:
                if len(item)>len(new_tuples[index_2]):
                    remove_index.append(index_2) #indefoirx
                elif len(item)<len(new_tuples[index_2]):
                    remove_index.append(index_1) #index
                elif len(item)==len(new_tuples[index_2]):
                    remove_index.append(index_2) #index
    remain_sets = set(range(len(new_tuples))).difference(set(remove_index))
    ##step 4, spit out final index and length
    index_1 = [] ## [0,1,2,3,3,3,4,4]
    index_length = []
    for i in remain_sets:
        index_1.append(dict_tuples[i])
    counts = Counter(index_1) ##this is a dictionary return Counter({3:3, 4:2, 0:1, 1:1, 2:1}) ##index:length
    list_counts = counts.most_common() ## convert to a list [(3,3), (4,2), (0,1), (1,1), (2,1)]
    for i in range(len(list_counts)):
        index_length.append([list_counts[i][0], list_counts[i][1]])
    index_length = sorted(index_length, key = itemgetter(0))
    return index_length

def search_func_groups(smiles): ##this is to search functional groups and print out them with numbers
    ## from https://github.com/curieshicy/JRgui/
    smarts = ["[$([CX2H0](=*)=*)]", "[$([CX2H1]#[!#7])]", "[$([CX2H0]#[!#7])]", "[OX2H]-[C]=O", "[#6X3H0;!$([#6X3H0](~O)(~O)(~O))](=[#8X1])[#8X2H0]",
              "[$([#6X3H0](=[OX1]));!$([#6X3](=[#8X1])~[#8X2]);R]=O", "[CH;D2;$(C-!@C)](=O)", "[OX2H;!$([OX2H]-[#6]=[O]);!$([OX2H]-a)]", "[O;H1;$(O-!@c)]", 
              "[#8X2H0;R;!$([#8X2H0]~[#6]=[#8])]", "[$([CX3H0](=[OX1]));!$([CX3](=[OX1])-[OX2]);!R]=O", "[OX2H0;!R;!$([OX2H0]-[#6]=[#8])]", 
              "[$([#7X3,#7X3+][!#8])](=[O])~[O-]", "[OX1H0;!$([OX1H0]~[#6X3]);!$([OX1H0]~[#7X3]~[#8])]", "[#7X2H0;R]",  "[#7X3H1;R]", "[#7X2H1]",
              "[#7X2H0;!R]","[#6X2]#[#7X1H0]","[NX3H2]", "[NX3H1;!R]", "[#7X3H0;!$([#7](~O)~O)]","[SX2H]","[#16X2H0;!R]","[#16X2H0;R]",  "[R;CX3H1,cX3H1]", 
              "[$([R;#6X3H0]);!$([R;#6X3H0]=[#8])]","[R;CX4H2]","[R;CX4H]","[R;CX4H0]", "[CX3H2]", "[!R;CX3H1;!$([CX3H1](=O))]",
              "[$([!R;#6X3H0]);!$([!R;#6X3H0]=[#8])]","[CX4H3]","[!R;CX4H2]", "[!R;CX4H]","[!R;CX4H0]","[F]","[Cl]","[Br]", "[I]"]
    tuples = []
    index_list = []
    final_index_and_length = []
    m = Chem.MolFromSmiles(str(smiles))    
    for index, smart in enumerate(smarts):
        if m.HasSubstructMatch(Chem.MolFromSmarts(smart)) == True: 
            tuples.append(m.GetSubstructMatches(Chem.MolFromSmarts(smart))) ## this is atom position
            index_list.append(index)
    temp = return_non_duplicate_index(tuples) # [[0, 1], [1, 1], [3, 1], [4, 7], [5, 6], [6, 1], [7, 1], [8, 1], [9, 1]]
    for i in temp:
        final_index_and_length.append([index_list[i[0]], i[1]])
    return final_index_and_length      

def compute_phys_properties(smiles):
    ## from https://github.com/curieshicy/JRgui/
    ## method from: K. G. Joback, R. C. Reid, ESTIMATION OF PURE-COMPONENT PROPERTIES FROM GROUP-CONTRIBUTIONS. Chemical Engineering Communications 57, 233-243 (1987).
    ## this doesn't look very accurate, but it's a start
    ##[[], [], ...[]] in total 41 nested list inside a list
    DB =    [[0.0026,	0.0028,	36,	26.15,	17.78,	142.14,	136.70,	2.74E+1,	-5.57E-2,	1.01E-4,	-5.02E-8,	4.720,	2.661,	None,	None],
            [0.0027,	-0.0008,46,	9.20,	-11.18,	79.30,	77.71,	2.45E+1,	-2.71E-2,	1.11E-4,	-6.78E-8,	2.322,	1.155,	None,	None],
            [0.0020,	0.0016,	37,	27.38,	64.32,	115.51,	109.82,	7.87,	         2.01E-2,	-8.33E-6,	1.39E-9,	4.151,	3.302,	None,	None],
            [0.0791,	0.0077,	89,	169.09,	155.50,	-426.72,-387.87,2.41E+1,	4.27E-2,	8.04E-5,       -6.87E-8,	11.051,	19.537,	1317.23,-2.578],
            [0.0481,	0.0005,	82,	81.10,	53.60,	-337.92,-301.95,2.45E+1,	4.02E-2,	4.02E-5,	-4.52E-8,	6.959,	9.633,	483.88,	-0.966],
            [0.0284,	0.0028,	55,	94.97,	75.97,	-164.50,-126.27,3.04E+1,	-8.29E-2,	2.36E-4,       -1.31E-7,	None,	6.645,	None,	None],
            [0.0379,	0.0030,	82,	72.24,	36.90,	-162.03,-143.48,3.09E+1,	-3.36E-2,	1.60E-4,       -9.88E-8,	3.197,	9.093,	740.92,	-1.713],
            [0.0741,	0.0112,	28,	92.88,	44.45,	-208.04,-189.20,2.57E+1,	-6.91E-2,	1.77E-4,	-9.88E-8,	2.406,	16.826,	2173.72,-5.057],
            [0.0240,	0.0184,	-25,	76.34,	82.83,	-221.65,-197.37,-2.81,	         1.11E-1,	-1.16E-4,	4.94E-8,	4.490,	12.499,	3018.17,-7.314],
            [0.0098,	0.0048,	13,	31.22,	23.05,	-138.16,-98.22,	1.22E+1,	-1.26E-2,	6.03E-5,	-3.86E-8,	5.879,	4.682,	440.24,	-0.953],
            [0.0380,	0.0031,	62,	76.75,	61.20,	-133.22,-120.50,6.45,	         6.70E-2,	-3.57E-5,	2.86E-9,	4.189,	8.972,	340.35,	-0.350],
            [0.0168,	0.0015,	18,	22.42,	22.23,	-132.22,-105.00,2.55E+1,	-6.32E-2,	1.11E-4,       -5.48E-8,	1.188,	2.410,	122.09,	-0.386],
            [0.0437,	0.0064,	91,	152.54,	127.24,	-66.57,	-16.83,	2.59E+1,	-3.74E-3,	1.29E-4,	-8.88E-8,	9.679,	16.738,	None,	None],
            [0.0143,	0.0101,	36,	-10.50,	2.08,	-247.61,-250.83,6.82,	        1.96E-2,	1.27E-5,	-1.78E-8,	3.624,	5.909,	675.24,	-1.340],
            [0.0085,	0.0076,	34,	57.55,	68.40,	55.52,	79.93,	8.83,	      -3.84E-3,	         4.35E-5,	-2.60E-8,	3.649,	6.528,	None,	None],
            [0.0130,	0.0114,	29,	52.82,	101.51,	31.65,75.61,1.18E+1,	        -2.30E-2,	1.07E-4,	-6.28E-8,	7.490,	6.930,	None,	None],
            [None,  	None,	None,	83.08,	68.91,	93.70,	119.66,	5.69,	       -4.12E-3,	1.28E-4,	-8.88E-8,	None,	12.169,	None,	None],
            [0.0255,	-0.0099,None,	74.60,	None,	23.61,	None,     None,	         None,	         None,	          None, 	None,	3.335,	None,	None],
            [0.0496,	-0.0101,91,	125.66,	59.89,	88.43,	89.22,	3.65E+1,	-7.33E-2,	1.84E-4,	-1.03E-7,	2.414,	12.851,	None,	None],
            [0.0243,	0.0109,	38,	73.23,	66.89,	-2.02,	14.07,2.69E+1,	        -4.12E-2,	1.64E-4,        -9.76E-8,	3.515,	10.788,	None,	None],
            [0.0295,	0.0077,	35,	50.17,	52.66,	53.47,	89.39,-1.21,	        7.62E-2,	-4.86E-5,	1.05E-8,	5.009,	6.436,	None,	None],
            [0.0169,	0.0074,	9,	11.74,	48.84,	123.34,	163.16,-3.11E+1,	2.27E-1,	-3.20E-4,	1.46E-7,	4.703,	1.896,	None,	None],
            [0.0031,	0.0084,	63,	63.56,	20.09,	-17.33,	-22.99,	3.53E+1,	-7.58E-2,	1.85E-4,	-1.03E-7,	2.360,	6.884,	None,	None],
            [0.0119,	0.0049,	54,	68.78,	34.40,	41.87,	33.12,	1.96E+1,	-5.61E-3,	4.02E-5,	-2.76E-8,	4.130,	6.817,	None,	None],
            [0.0019,	0.0051,	38,	52.10,	79.93,	39.10,	27.76,	1.67E+1,	4.81E-3,	2.77E-5,	-2.11E-8,	1.557,	5.984,	None,	None],
            [0.0082,	0.0011,	41,	26.73,	8.13,	2.09,	11.30,	-2.14,	        5.74E-2,	-1.64E-6,	-1.59E-8,	1.101,	2.544,	259.65,	-0.702],
            [0.0143,	0.0008,	32,	31.01,	37.02,	46.43,	54.05,	-8.25,	        1.01E-1,	-1.42E-4,	6.78E-8,	2.394,	3.059,	-245.74,0.912],
            [0.0100,	0.0025,	48,	27.15,	7.75,	-26.80,	-3.68,	-6.03,	         8.54E-2,	-8.00E-6,	-1.80E-8,	0.490,	2.398,	307.53,	-0.798],
            [0.0122,	0.0004,	38,	21.78,	19.88,	8.67,	40.99,	-2.05E+1,	1.62E-1,	-1.60E-4,	6.24E-8,	3.243,	1.942,	-394.29,1.251],
            [0.0042,	0.0061,	27,	21.32,	60.15,	79.72,	87.88,	-9.09E+1,	5.57E-1,	-9.00E-4,	4.69E-7,	-1.373,	0.644,	None,	None],
            [0.0113,	-0.0028,56,	18.18,	-4.32,	-9.630,	3.77,	2.36E+1,	-3.81E-2,	1.72E-4,	-1.03E-7,	-0.473,	1.724,	495.01,	-1.539],
            [0.0129,	-0.0006,46,	24.96,	8.73,	37.97,	48.53,	-8.00,	        1.05E-1,	-9.63E-5,	3.56E-8,	2.691,	2.205,	82.28,	-0.242],
            [0.0117,	0.0011,	38,	24.14,	11.14,	83.99,	92.36,	-2.81E+1,	2.08E-1,	-3.06E-4,	1.46E-7,	3.063,	2.138,	None,	None],
            [0.0141,	-0.0012,65,	23.58,	-5.10,	-76.45,	-43.96,	1.95E+1,	-8.08E-3,	1.53E-4,	-9.67E-8,	0.908,	2.373,	548.29,	-1.719],
            [0.0189,	0.0000,	56,	22.88,	11.27,	-20.64,	8.42,	-9.09E-1,	9.50E-2,	-5.44E-5,	1.19E-8,	2.590,	2.226,	94.16,	-0.199],
            [0.0164,	0.0020,	41,	21.74,	12.64,	29.89,	58.36,	-2.30E+1,	2.04E-1,	-2.65E-4,	1.20E-7,	0.749,	1.691,	-322.15,1.187],
            [0.0067,	0.0043,	27,	18.25,	46.43,	82.23,	116.02,	-6.62E+1,	4.27E-1,	-6.41E-4,	3.01E-7,	-1.460,	0.636,	-573.56,2.307],
            [0.0111,	-0.0057,27,	-0.03,	-15.78,	-251.92,-247.19,2.65E+1,	-9.13E-2,	1.91E-4,	-1.03E-7,	1.398,	-0.670,	None,	None],
            [0.0105,	-0.0049,58,	38.13,	13.55,	-71.55,-64.31,	3.33E+1,	-9.63E-2,	1.87E-4,	-9.96E-8,	2.515,	4.532,	625.45,	-1.814],
            [0.0133,	0.0057,	71,	66.86,	43.43,	-29.48,	-38.06,	2.86E+1,	-6.49E-2,	1.36E-4,	-7.45E-8,	3.603,	6.582,	738.91,	-2.038],
            [0.0068,	-0.0034,97,	93.84,	41.69,	21.06,	5.74,	3.21E+1,	-6.41E-2,	1.26E-4,	-6.87E-8,	2.724,	9.520,	809.55,	-2.224]] 

    mol = Chem.MolFromSmiles(str(smiles))
    NoA = Chem.AddHs(mol).GetNumAtoms()
    MW = Descriptors.MolWt(Chem.AddHs(mol))
    LogP = Descriptors.MolLogP(Chem.AddHs(mol))
    MR = Descriptors.MolMR(Chem.AddHs(mol))

    double_lists = search_func_groups(smiles)

    entry_index_by_users = []
    entry_data_by_users = []

    for item in double_lists:
        entry_index_by_users.append(item[0])
        entry_data_by_users.append(item[1])
        
    fiveteen_columns = [] ##length  = 15*len(entry_index_by_users)
    for index, data in zip(entry_index_by_users, entry_data_by_users):
        for i in range(15):
            if DB[index][i] == None:
                temp = None
            else:
                temp = data*DB[index][i]
            fiveteen_columns.append(temp)

    temperature = 310
    Tc = []
    Pc = []
    Vc = []
    Tb = []
    Tm = []
    Hfor = []
    Gf = []
    Cpa = []
    Cpb = []
    Cpc = []
    Cpd = []
    Hfus = []
    Hvap = []
    Ya = []
    Yb =[]        
    fc = fiveteen_columns ## short hand
    for i in range(len(entry_index_by_users)):
        Tc.append(fc[i*15])
        Pc.append(fc[i*15 + 1])
        Vc.append(fc[i*15 + 2])
        Tb.append(fc[i*15 + 3])
        Tm.append(fc[i*15 + 4])
        Hfor.append(fc[i*15 + 5])
        Gf.append(fc[i*15 + 6])
        Cpa.append(fc[i*15 + 7])
        Cpb.append(fc[i*15 + 8])
        Cpc.append(fc[i*15 + 9])
        Cpd.append(fc[i*15 + 10])
        Hfus.append(fc[i*15 + 11])
        Hvap.append(fc[i*15 + 12])
        Ya.append(fc[i*15 + 13])
        Yb.append(fc[i*15 + 14])
    try:
        BoilingPoint = 198.2 + sum(Tb)
    except:
        BoilingPoint = None        
    try:
        MeltingPoint = 122.5 + sum(Tm)
    except:
        MeltingPoint = None
    try:
        CriticalTemp =  (sum(Tb) + 198.2)/(0.584 + 0.965*sum(Tc) - sum(Tc)**2)
    except:
        CriticalTemp = None
    try:
        CriticalPress = 1./(0.113 + 0.0032*float(NoA) - sum(Pc))**2
    except:
        CriticalPress = None
    try:
        CriticalVolume = 17.5 + sum(Vc)
    except:
        CriticalVolume = None
    try:
        EnthalpyForm = 68.29 + sum(Hfor)
    except:
        EnthalpyForm = None
    try:
        GibbsEnergy = 53.88 + sum(Gf)
    except:
        GibbsEnergy = None
    try:
        HeatCapacity = (sum(Cpa) - 37.93) + (sum(Cpb) + 0.210)*float(temperature) + (sum(Cpc) - 3.91*10**(-4))*float(temperature)**2 + (sum(Cpd) + 2.06*10**(-7))*float(temperature)**3
    except:
        HeatCapacity = None
    try:
        EnthalpyVap = 15.30 + sum(Hvap)
    except:
        EnthalpyVap = None
    try:
        EnthalpyFus = -0.88 + sum(Hfus)
    except:
        EnthalpyFus = None
    try:
        LiquidVisco = float(MW)*math.exp((sum(Ya) - 597.82)/float(temperature) + sum(Yb) - 11.202)
    except:
        LiquidVisco = None
    try:
        CrystalSolub_1 = 10**(0.8 - float(LogP) - 0.01*(sum(Tm)+122.5 - 273.15 - 25.))*1000.*float(MW)
    except:
        CrystalSolub_1 = None
    try:
        CrystalSolub_2 = 10**(0.5 - float(LogP) - 0.01*(sum(Tm)+122.5 - 273.15 - 25.))*1000.*float(MW)
    except:
        CrystalSolub_2 = None
    try:
        AmorphSolub_1 = 10**(0.8 - float(LogP) - 0.01*(sum(Tm)+122.5 - 273.15 - 25.)) *1000.*float(MW)*math.exp((sum(Hfus)-0.88)*(sum(Tm) + 122.5 - float(temperature))*float(temperature)/(sum(Tm) + 122.5)**2/(2.479*float(temperature)/298.))
    except:
        AmorphSolub_1 = None
    try:
        AmorphSolub_2 = 10**(0.5 - float(LogP) - 0.01*(sum(Tm)+122.5 - 273.15 - 25.)) *1000.*float(MW)*math.exp((sum(Hfus)-0.88)*(sum(Tm) + 122.5 - float(temperature))*float(temperature)/(sum(Tm) + 122.5)**2/(2.479*float(temperature)/298.))
    except:
        AmorphSolub_2 = None
    return MeltingPoint-273.15

def string2mp(name, namespace='name', local_only=False, include_predictions=True):
    mp, mp_origin = None, None
    # try the chemicals package
    if is_cas(name):
        cas = name
        if pd.isna(mp):
            mp = chemicals.Tm(cas)
            if mp: mp = mp-273.15
            methods = chemicals.Tm_methods(cas)
            if methods:
                mp_origin = 'chem/'+methods[0]
                # Keep Joback as a last-resort estimate rather than a preferred local MP.
                if methods[0] == 'JOBACK':
                    mp, mp_origin = None, None
        # experimental values
        if pd.isna(mp):
            # some rows have multiple comma-separated cas numbers
            mask = dfmp_expt['CAS'].str.contains(cas)
            if sum(mask):
                subdf = dfmp_expt[mask]['CAS'].str.split(', ')
                for index,subcas in zip(subdf.index,subdf):
                    if cas in subcas:
                        mp = float(dfmp_expt.iloc[index]['MP'])
                        mp_origin = 'expt'
        # predicted values from Mansouri
        #if pd.isna(mp):
        #    mask = dfmp_pred['Substance_CASRN'] == cas
        #    if sum(mask):
        #        mp = float(dfmp_pred[mask]['NCCT_MP'])
        #        mp_origin = 'pred'
        # predicted values from OPERA (CompTox dashboard)
        if include_predictions and pd.isna(mp):
            mask = df_pred['CASRN'] == cas
            if sum(mask):
                mp = float(df_pred[mask]['MELTING_POINT_DEGC_OPERA_PRED'].iloc[0])
                mp_origin = 'comptox/pred'
    if local_only:
        return mp, mp_origin
    # try to scrape from PubChem
    if pd.isna(mp):
        content = None
        try:
            compounds = pcp.get_compounds(name, namespace=namespace)
            c = compounds[0]
            cid = c.cid
            url = f'https://pubchem.ncbi.nlm.nih.gov/rest/pug_view/data/compound/{cid}/JSON'
            #fid = urllib.request.urlopen(url)
            #webpage = fid.read().decode('utf-8')
            webpage = requests.get(url).text
            content = json.loads(webpage)
        except:
            pass
        if content:
            mp_list = []
            for i in content['Record']['Section']:
                if i['TOCHeading'] == 'Chemical and Physical Properties':
                    for j in i['Section']:
                        if j['TOCHeading'] == 'Experimental Properties':
                            for k in j['Section']:
                                if k['TOCHeading'] == 'Melting Point':
                                    for ii in k['Information']:
                                        try:
                                            mp_string = ii['Value']['StringWithMarkup'][0]['String']
                                            #rho_string = rho_string.replace('Relative density (water = 1): ', '')
                                            #print(mp_string)
                                            #tmp_rho = re.match('(?:\d+(?:\.\d*)?|\.\d+)',rho_string)
                                            #m = re.match('((?:\d+(?:\.\d*)?|\.\d+))(?:-((?:\d+(?:\.\d*)?|\.\d+)))?',rho_string)
                                            m = re.match(r'(-?(?:\d+(?:\.\d*)?|\.\d+))(?:-((?:\d+(?:\.\d*)?|\.\d+)))?( ?°?C)',mp_string)
                                            if m is not None:
                                                groups = m.groups()
                                                if len(groups):
                                                    for g in groups:
                                                        try:
                                                            tmp_mp = float(g)
                                                            mp_list.append(tmp_mp)
                                                        except:
                                                            continue
                                            m = re.match(r'(-?(?:\d+(?:\.\d*)?|\.\d+))(?:-((?:\d+(?:\.\d*)?|\.\d+)))?( ?°?F)',mp_string)
                                            if m is not None:
                                                groups = m.groups()
                                                if len(groups):
                                                    for g in groups:
                                                        try:
                                                            # (T/F − 32) × 5/9 = T/C
                                                            tmp_mp = (float(g)-32)*5/9
                                                            mp_list.append(tmp_mp)
                                                        except:
                                                            continue
                                        except:
                                            continue
            if mp_list:
                ## remove outliers using interquartile range (IQR)
                mp_list = np.array(mp_list)
                q75,q25 = np.percentile(mp_list,[75,25])
                intr_qr = q75-q25
                hi = q75+(1.5*intr_qr)
                lo = q25-(1.5*intr_qr)
                mask = (mp_list <= hi) & (mp_list >= lo)
                mp_list = mp_list[mask]
                mp = np.mean(mp_list)
                mp_origin = 'pubchem'
            else:
                mp, mp_origin = None, None
    # try to scrape from DSSTOX website...
    if try_dsstox:
        if pd.isna(mp):
            dtxsid = None
            # try to find it via the dsstox dashboard
            try:
                name_urlsafe = urllib.parse.quote(name)
                url = f'https://comptox.epa.gov/dashboard/search-results?input_type=synonym_substring&inputs={name_urlsafe}'
                fid = urllib.request.urlopen(url)
                webpage = fid.read().decode('utf-8')
                hits = re.findall(r'DTXSID[0-9]+', webpage)
                if len(hits):
                    dtxsid = hits[0]
            except:
                pass
            if dtxsid:
                mysoup = None
                url = f'https://comptox.epa.gov/dashboard/chemical/properties/{dtxsid}'
                #print(url)
                try:
                    #driver = selenium.webdriver.Firefox()
                    #driver.set_page_load_timeout(15)
                    #driver.get(url)
                    #driver_exe = 'chromedriver'
                    #driver_exe = chromedriver_binary.chromedriver_filename
                    options = Options()
                    options.add_argument("--headless") # runs in background instead of showing browser window
                    service = Service(driver_exe)
                    driver = selenium.webdriver.Chrome(service=service, options=options)
                    #driver = selenium.webdriver.Chrome(driver_exe, options=options)
                    driver.set_page_load_timeout(15)
                    driver.get(url)
                    webpage = driver.page_source
                    driver.quit()
                    mysoup = bs4.BeautifulSoup(webpage, features='lxml')
                except:
                    pass
                    #print("timeout")
                # column of property names
                if mysoup:
                    ifound = None
                    rows = mysoup.find_all('div', attrs={'col-id':'property'})
                    for i,row in enumerate(rows):
                        if 'Melting Point' in row.text:
                            ifound = i
                            break
                    if ifound:
                        rows = mysoup.find_all('div', attrs={'col-id':'exavg'})
                        text = rows[ifound].text
                        value = re.sub(r' \([0-9]*\)', '', text.strip())
                        try:
                            mp = float(value)
                            mp_origin = 'dsstox/expt'
                        except:
                            mp, mp_origin = None, None
                        if include_predictions and pd.isna(mp):
                            rows = mysoup.find_all('div', attrs={'col-id':'predavg'})
                            text = rows[ifound].text
                            value = re.sub(r' \([0-9]*\)', '', text.strip())
                            try:
                                mp = float(value)
                                mp_origin = 'dsstox/pred'
                            except:
                                mp, mp_origin = None, None
                    else:
                        mp, mp_origin = None, None
            else:
                mp, mp_origin = None, None
    if pd.isna(mp): mp, mp_origin = None, None
    return mp, mp_origin

def smiles2mp(smiles):
    try:
        SUPPORTED_ATOM_SET = {6, 7, 8, 9, 16, 17, 35, 53}
        m = Chem.MolFromSmiles(str(smiles))
        atom_num_set = set([a.GetAtomicNum() for a in m.GetAtoms()])
        if atom_num_set.issubset(SUPPORTED_ATOM_SET):
            mp = compute_phys_properties(smiles)
        else:
            mp = None
    except:
        mp = None
    return mp

def smiles2mp_opera(smiles):
    descs = padelpy.from_smiles(smiles, descriptortypes='mp/descriptors.xml')
    #dfd = pd.DataFrame(descs,index=[0])
    #dfd = dfd.replace('',0).infer_objects(copy=False)
    #dfd = pd.DataFrame(dfd, dtype=float)
    dfd = pd.DataFrame(descs, index=[0]).apply(pd.to_numeric, errors="coerce").fillna(0.0).astype(float)
    X = np.array(dfd[my_opera_data_mp.desc_list])
    X_scale = my_opera_data_mp.scaler_X.transform(X)
    y_pred = my_opera_data_mp.scaler_y.inverse_transform(my_opera_data_mp.knn_all.predict(X_scale))
    return y_pred[0][0]

def mol2mp(cas, name, smiles):
    mp, mp_origin = None, None
    # Prefer MP values already available locally before calculating or making network requests.
    if pd.isna(mp) and cas:
        try:
            mp, mp_origin = string2mp(cas, local_only=True)
        except:
            mp, mp_origin = None, None
    if pd.isna(mp) and smiles:
        # OPERA is optional; it can fail when Java/PaDEL is unavailable.
        try:
            mp = smiles2mp_opera(smiles)
            mp_origin = 'opera/calc'
        except:
            mp, mp_origin = None, None
    if pd.isna(mp) and cas:
        try:
            mp, mp_origin = string2mp(cas)
        except:
            mp, mp_origin = None, None
    if pd.isna(mp) and name:
        try:
            mp, mp_origin = string2mp(name)
        except:
            mp, mp_origin = None, None
    if pd.isna(mp) and smiles:
        try:
            mp = smiles2mp(smiles)
            mp_origin = 'joback-reid/calc'
        except:
            mp, mp_origin = None, None
    return mp, mp_origin

def getLogP(cas,mol):
    LogP, LogP_origin = None, None
    if cas:
        mask = df_pred['CASRN'] == cas
        if sum(mask):
            LogP = float(df_pred[mask]['OCTANOL_WATER_PARTITION_LOGP_OPERA_PRED'])
            LogP_origin = 'comptox/pred'
    if pd.isna(LogP) and mol:
        LogP = Crippen.MolLogP(mol)
        LogP_origin = 'rdkit/calc'
    return LogP, LogP_origin