Artificial Intelligence Used to Study NBA Players Can Help Predict COVID-19 Variant Trends, Trajectories

Pharmacy Times interviewed Dong Xu, PhD, MS, curators' distinguished professor, Department of Electrical Engineering and Computer Science, University of Missouri College of Engineering, on research assessing the application of a form of artificial intelligence (AI) to help scientists develop new drug therapies for medical treatments targeting cancers and other diseases.

Alana Hippensteele: Hi, I’m Alana Hippensteele with Pharmacy Times. Joining me is Dong Xu, PhD, MS, a curators' distinguished professor in the Department of Electrical Engineering and Computer Science at the University of Missouri College of Engineering, who is here to discuss research that is looking to apply a form of artificial intelligence, that was previously used to analyze how NBA players move their bodies, to now help scientists develop new drug therapies for medical treatments targeting cancers and other diseases.

What have been the results of the research conducted using this AI thus far in terms of potential for new drug therapies?

Dong Xu: Yeah, drug development is a really long process. It has many, many steps, as I mentioned. First of all, you decide what is the target, and then what kind of drug, and also study side effects. So our approach can provide a general framework for part of this study.

So, for example, we just studied 3 proteins. Some of them actually directly relate to cancer. Hopefully, our study results can provide a basis for other people to further study these proteins to understand the cancer process, and then with it a better understanding of what mechanism probably can help develop some new therapies or some strategies.

Alana Hippensteele: What are some opportunities the use of this AI in drug development makes possible that might be on the horizon in terms of future research avenues?

Dong Xu: Yeah. So I think one possible strategy, as I mentioned, is not directly targeting functional sites. For example, in the old days, I mean, so far, pretty much, people primarily focus on so-called active sites. So the target positions are relatively limited.

With this method, it opens the door to other potential sites. For example, you could have a pass on the protein’s surface. So along the pass, multiple positions could be targeted. So we hope this method would allow people to explore more potential drug target sites, so that I believe could be useful.

Alana Hippensteele: Are there any other areas within health care research where this technology might also be applicable outside of the drug therapy focus?

Dong Xu: Yeah, so that's an interesting question. In fact, this method potentially can be used for some other things. So, to be more technically specific, it is called the neuro relational inference. That means that you actually can infer future trends. So what we're doing right now, actually, is to apply this method to predict COVID-19 development trends. So that is, in different areas, we use one country or one state as a unit, and then we see the relationship between those counties in terms of whether they are neighbors, and then we have a time course of COVID-19. In each country, or each state, we also have different variants, like the omicron variant, etc, and then we have a trajectory.

So we are using this same method, basically, to build a graph. In this case, each node will be either a state or a county, and then the edges between nodes will be the neighboring relationship with just a simple distance. Then each node will have a time course, and that's how we see the COVID-19 up and down, and then we can use methods to project let's say, what would happen one week from now or one month from now, what kind of variants will change over time? So that's another application of using these neural relational inference for predicting trajectories.

Then potentially can be used to predict other dynamic trajectories, for example, the mutation trajectory and also other cellular change trajectories. So we hope we can use this method to study more biological topics.

Alana Hippensteele: That's very interesting. Do you have any future plans in terms of what you will be researching on the horizon?

Dong Xu: Yeah, so in fact, my current focus, starting from about 10 years ago, is to apply deep learning for biomedical studies. So right now, we study a lot of single-cell data analysis, that is data derived from an individual cell in terms of gene expression, in terms of so-called regulatory changes, and then we infer from single-cell data some useful information related to, let's say, the immune system, cancers, and others. So we use a lot of deep learning for that in our study.

We also do other studies, for example, to predict where protein will be localized, or a so-called putting localization problem using deep learning. So deep learning, just like its role in many other fields, really is playing a very important role in biomedical research.

What we see with deep learning is there are a lot of things which couldn't be done without deep learning. Now, it becomes possible to use deep learning to explore and also many other problems. There were methods [before], but with deep learning, we do see significant improvement for those methods. So I probably will continue to focus on the application of deep learning in biomedical research for the next 5 to 10 years.