Using Deep Convolutional Neural Networks to predict compound-target interactions

Getting a new drug from the laboratory into the market is a lengthy and costly process which takes on average 12 years and over US$350 million to accomplish. It is composed of roughly 3 phases: (1) pre-clinical research, (2) clinical studies, and (3) the new drug application review. In this work, we propose an artificial intelligence system which will shorten the time it takes for pharmaceutical companies to identify novel drugs (compounds) for a given target (usually a protein or a protein complex). Our proposed system will predict the interactions between compounds and targets based on their 3D structure, and will be based on a large-scale and publically available database (ChEMBL) which contains information on 11,019 targets and more than 1.5 million compounds. We expect that our “learning from examples” approach, compared to the more traditional approach of manual engineering, will allow our system to better “understand” the structure of compounds and targets, and therefore, better predict their interactions.

Intern: 
Rotem Golan
Faculty Supervisor: 
Christian Jacob
Project Year: 
2016
Province: 
Alberta
Partner: 
Discipline: 
Program: