ChEMBL Multitask Nerual Network model

Small and fast target prediction model trained on a panel of targets using ChEMBL data. The model can be used in off-target prediction scenarios with large collections of compounds.

• Based on the blogpost: http://chembl.blogspot.com/2019/05/multi-task-neural-network-on-chembl.html
• Model available in KNIME thanks to Greg Landrum: https://www.knime.com/blog/interactive-bioactivity-prediction-with-multitask-neural-networks

The model is exported to the ONNX format so it can be used in any programming language able to generate fingerprints with RDKit

Data Extraction

Activities in ChEMBL with the following requirements are extracted

• activities.standard_units = 'nM'
• activities.standard_type IN ('EC50', 'IC50', 'Ki', 'Kd', 'XC50', 'AC50', 'Potency')
• activities.data_validity_comment IS NULL
• activities.standard_relation IN ('=', '<')
• activities.potential_duplicate = 0 AND assays.confidence_score >= 8
• target_dictionary.target_type = 'SINGLE PROTEIN'

Keeping targets

• with at least 100 active and 100 inactive compounds
• mentioned in at least 2 publications

Using IDG protein family activity thresholds

• Kinases: <= 30nM
• GPCRs: <= 100nM
• Nuclear Receptors: <= 100nM
• Ion Channels: <= 10μM
• Non-IDG Family Targets: <= 1μM

When multiple measurements for a target-pair are found, the one with the lowest concentration is selected. This intentionally biases the model toward sensitivity.

Example to predict in Python using the ONNX Runtime

import onnxruntime
import numpy as np
from rdkit import Chem
from rdkit.Chem import rdMolDescriptors

FP_SIZE = 1024
RADIUS = 2

def calc_morgan_fp(smiles):
    mol = Chem.MolFromSmiles(smiles)
    fp = rdMolDescriptors.GetMorganFingerprintAsBitVect(
        mol, RADIUS, nBits=FP_SIZE)
    a = np.zeros((0,), dtype=np.float32)
    Chem.DataStructs.ConvertToNumpyArray(fp, a)
    return a

def format_preds(preds, targets):
    preds = np.concatenate(preds).ravel()
    np_preds = [(tar, pre) for tar, pre in zip(targets, preds)]
    dt = [('chembl_id','|U20'), ('pred', '<f4')]
    np_preds = np.array(np_preds, dtype=dt)
    np_preds[::-1].sort(order='pred')
    return np_preds

# load the model
ort_session = onnxruntime.InferenceSession("trained_models/chembl_34_model/chembl_34_multitask.onnx", providers=['CPUExecutionProvider'])

# calculate the FPs
smiles = 'CN(C)CCc1c[nH]c2ccc(C[C@H]3COC(=O)N3)cc12'
descs = calc_morgan_fp(smiles)

# run the prediction
ort_inputs = {ort_session.get_inputs()[0].name: descs}
preds = ort_session.run(None, ort_inputs)

# example of how the output of the model can be formatted
preds = format_preds(preds, [o.name for o in ort_session.get_outputs()])

In Julia using RDKitMinimalLib.jl and ONNX.jl

import RDKitMinimalLib: get_mol, get_morgan_fp
import Umlaut: play!
import ONNX
import JSON

path = "chembl_31_multitask.onnx"
targets = JSON.parsefile("targets_31.json")

# dummy input
dummy = rand(Float32, 1024, 1)
# load the model
mt_chembl = ONNX.load(path, dummy)

# load molecule and calc morgan fingerprint
mol = get_mol("CC(=O)Oc1ccccc1C(=O)O")
fp_details = Dict{String, Any}("nBits" => 1024, "radius" => 2)
mfp = get_morgan_fp(mol, fp_details)

# convert the bitstring to a 1024×1 Matrix{Float32}
mfp = map(x->parse(Float32,string(x)),collect(mfp))
mfp = reshape(mfp, (length(mfp), 1))

# test a molecule
pred = play!(mt_chembl, mfp)
pred = collect(Iterators.flatten(pred))

res = tuple.(targets, pred)
res = sort(res, by=res->res[2], rev=true)

C++ REST microservice

https://github.com/eloyfelix/pistache_predictor

Try it online!

Using both RDKit Javascript MinimalLib and ONNX.js. Hosted in github pages: https://chembl.github.io/chembl_multitask_model