- Integrated Systems
- About Us
- Integrated Systems
- About Us
What we have learned from working on COVID-19
Join Dr. Goh at NVIDIA GTC21. He has been on the frontline of technology advancements and global consortia work on COVID-19, and will share specific proven methods to accelerate the drug discovery process.
The process of developing a drug or vaccine is long and complex. It was therefore vital to identify ways to accelerate or optimize the associated research and development work. To this end, HPE worked on projects with many of our customers in areas such as drug development, clinical trials, and research aggregation.
I would like to share some of the learnings we gleaned from this work at our GTC session (ID: S32009).
There are many paths to developing a drug.
In the case of COVID-19, one of the paths is to first determine the 3D structure of the COVID-19 virus spike protein. This is important because the shape of the virus spike determines how it attaches to the human ACE2 receptor as part of the infection process. The amount of attraction between the virus spike and our ACE2 receptors is known as binding energy. This binding energy is crucial because any drug that has a greater attraction to the virus spike may be used to block the virus from attaching itself to our ACE2 receptors.
The problem is, running high-fidelity computer simulations to test the attraction of each drug, from the massive library of available drugs candidates, with the virus is both expensive and time-consuming. Consequently, to short list drugs candidates before testing, we trained a machine learning model to identify drug candidates that were more likely to have the requisite binding energy. This allows the lengthy and costly simulations to be limited to the best candidates for a match.
Next, let us look at clinical trials.
One outcome in early clinical trials for vaccines is having fewest volunteers in the trial with adverse reactions and conversely most of the volunteers with good antibody responses. Our task was to determine if a blood sample from a volunteer could be used to predict whether that volunteer was likely to experience an adverse reaction, and whether that volunteer was likely to attain good antibody response.
To do this, we developed a set of machine learning models to look at gene expression data measured from the blood sample and see if they provided clues that would enable accurate predictions. Early last year, we began with anonymized data from a previous clinical trial of a different vaccine. In that trial, blood samples from volunteers had been gathered early. We also had eventual results showing for each volunteer whether there was adverse reactions and whether there was good antibody responses. We then worked on classifying these gene expression data into the specified outcomes. From the work already conducted, we have seen some important results that may contribute to optimizing clinical trials.
Another challenge associated with clinical research.
In the span of a year, the massive amount of work on COVID-19 resulted in more than 200,000 scholarly articles totaling millions of pages. That means many questions about COVID-19 have been explored and the results of these explorations are in those pages. The problem, in a sense, is that there is simply too much data. Even if the answer to a particular question is available somewhere in the research, finding it could be too time consuming.
To overcome this challenge, we created an ensemble natural language processing system called QnA. This system allows researchers to ask a complex domain-specific question using natural language—no need for any programming, or special operators, or anything else—and get ranked answers back from wherever they exist in this massive body of work. If you have used a search engine to find a very specific answer to a complex question, you know how hard it can be. The goal of QnA was to make it as simple as possible for clinicians and scientists to use productively.
This is where we are today. Our hope is that our work will help us, and our customers be more prepared in the future.
Learn more about applying AI and ML to accelerate drug discovery and optimize clinical trials at GTC21
Join HPE at NVIDIA GTC21 for a transformative global event that brings together creative minds looking to ignite ideas, build new skills, and forge new connections to take on our biggest challenges. It all comes together online April 12 – 16. For more information on registering for this and other sessions that HPE is giving, please visit hpe.com/events/gtc .
Eng Lim Goh
Senior Vice President & Chief Technology Officer, AI
Hewlett Packard Enterprise
Dr. Eng Lim Goh is senior vice president and chief technology officer for artificial intelligence at HPE. As principal investigator of the experiment aboard the International Space Station to operate autonomous supercomputers on long duration space travel, Dr. Goh was awarded NASA’s Exceptional Technology Achievement Medal. In addition to co-inventing blockchain-based swarm learning applications, he oversees deployment of artificial intelligence to Formula 1 racing, works on industrial application of technologies behind a champion poker bot, co-designed the systems architecture for simulating a biologically detailed mammalian brain, and leads a team on machine learning of gene expression data from vaccine clinical trial. He has been granted nine U.S. patents, with four others pending.