Overview / Usage

COVID-19 is a growing global threat already resulting in thousands of deaths world-wide. There are currently no effective drugs or vaccines for treating COVID-19. We are developing and applying computational methods to rapidly identify potential drug candidates for treating COVID-19.

COVID-19 is the most dangerous viral outbreak since the 2009 H1N1 influenza pandemic. The viral agent, SARS-CoV-2, is closely related to the SARS virus responsible for a 2002 pandemic. COVID-19 is less lethal but more infectious than SARS. Even at this early stage of a global outbreak, COVID-19 has already killed three times more people than the entire SARS pandemic.

Unfortunately, no drug or vaccines are available to treat human COVID-19. Research efforts are underway globally to develop therapies. However, new therapies will require extensive experimentation that will require months to years to perform. To accelerate drug discovery in the face of a global emergency, we are adopting a two-pronged strategy:

1. We are focusing on investigating clinically used and well-studied experimental drugs, including known antiviral drugs, for potential activity against SARSCoV-2. These drugs have a higher chance of being effective in humans without toxicity or dangerous side effects.

2. We are developing and using new machine learning strategies for generating and evaluating drug candidate molecules.

Methodology / Approach

Our targeted approach currently focuses on the SARS-CoV-2 “main protease” enzyme. This enzyme is necessary for viral replication. Drugs that block a similar enzyme are a key component of antiHIV therapy. However, the SARS-CoV-2 main protease differs from the HIV protease and will probably require different chemistries for effective inhibition.

Our lab is also developing deep learning methods for drug chemistry research. We have been highly successful in developing new predictive strategies of drug effectiveness as represented by top finishes in blinded competitions such as the Open Source Malaria and Drug Design Data Resource challenges. In the second stage of our research, we will be applying generative machine learning strategies such as generative adversarial networks and variational autoencoders for the in silico proposal and evaluation of new drug molecules.

Technologies Used

• Python
• Machine Learning
• MKL
• Intel CPUs
• Intel scikit-learn
• Generative Adversarial Networks

Hardware used - Devcloud ( Intel® Xeon® Gold 6128 CPUs Intel® 9Gen, Graphics NEO)

Other links

• Ho-Leung Ng Protein Structure Lab

Collaborators