Abstract

Author: [email protected] (Any bug report is welcome)

Time Created: Aug 2016

Time Updated: Dec 2016

Addr: Shenzhen, China

Description: We attempt to explore ChEMBL's Inhibitors by deep neural network

Website: https://xiaotaw.github.io/chembl/

Background

(add background for using DNN and RF to build this qsar model)

Problem

(add one sentence abstract for current challenge)

Solution

(how we solve the problem)

Method

1 get data

 1.1 positive dataset was downloaded from chembl database
 1.2 negtive dataset was selected from pubchem and chembl database(based on a reasonable assumption that almost the compound in pubchem was NOT the substrate of a protein kinase)

2 build the model

 2.1 deep neural network(based on tensorflow)
 2.2 random forest(based on scikit-learn)
 2.3 a 'Tree' comprises one 'Term' and several 'Branches', where the 'Term' extracts the mutual figures of all the protein kinase.

3 train and evaluation

 3.1 we train the model seperately and jointly, and then apply the model on pubchem dataset for virtual screening.