AI Used for Analysis of NBA Players’ Movement May Help Develop Cancer Treatments

Dong Xu, PhD, MS, curators' distinguished professor at the University of Missouri College of Engineering, discusses how artificial intelligence is being used to develop new drug therapies for medical treatments targeting cancers and other diseases.

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.

So Dr. Xu, what is the type of AI the University of Missouri is using in this drug development research, and how is it able to help scientists with this research?

Dong Xu: Yeah, so it's one special type of AI. So AI is a very big umbrella, and then within that big umbrella, there's a branch called machine learning, I think most people probably heard about it. Then within the branch of machine learning, there's actually a new type of machine learning, it's called deep learning. Then within deep learning, there is one methodology called the graph neural network. So we use the graph neural network.

So the feature of graph neural network is basically to consider different identities and relationships. So in our case, we consider different atoms or amino acids, how they interact with each other, how the dynamic changes, and then we'll build the graph. Basically, in the graph, we have the node representing one amino acid, and then between 2 nodes represents the interaction between those 2 amino acids. So that's kind of the special method that we are using.

Alana Hippensteele: That's fascinating. How was this AI previously used to analyze how NBA players move their bodies, and what was the impact of that research?

Dong Xu: Yes, so we actually borrowed this methodology. For other areas, for example, they use this method to study NBA players. The way they do it is that they find some representative points on a player, and then to see between those points what kind of original patterns [there are], and then they can derive a lot of information’s based on the dynamics of this playing.

For example, they can tell which hand they really use to carry a ball, and also the performance and how different players coordinate with each other. So it is actually very informative to infer from a video without human intervention, and then you input a video, the output is some analysis of this play.

Then this has been used for some other analysis, for example, in astronomy or in the atomic world, [looking at] how they interact with each other. But nobody has applied this method for biological studies. So we are the first one using it for studying protein.

Alana Hippensteele: That is very, very interesting. How can this AI help to develop new drug therapies for medical treatments targeting cancers?

Dong Xu: Yeah, so. So the way we do it, our input is the trajectory of the molecular dynamic simulation of protein. So this simulation on cancer has been in academia for many years. Win fact, when I was a PhD student I did it about 30 years ago to basically simulate how protein moves, but it's actually not easy to analyze search results. So, you see lots of movement, but you do not really see a lot of insight from the dynamics, and then our method really can provide some insight about the protein interaction mode. That is, between those amino acids, how they interact with each other, how they coordinate with each other.

In fact, in this paper, we studied a very special type of interaction called the allosteric effect that is in a protein—this kind of interaction is very interesting. So you have one site may interact with a drug or with another protein, but this interaction is affected by another site of a protein which is pretty far away. So it is very difficult to understand how these remote sites impact this allosteric effect.

So people have actually did a simulation to try to figure it out, but it's very difficult to figure out how that interaction [sic] [impacts the] protein, and this method we developed really can project how this remote site impacts this binding site. We can tell which amino acids are involved. So it becomes very useful to study this pathway. Based on that, we actually can develop an intervention strategy, for example, in the old days, people pretty much designed drugs by the binding site, but since we can find a pass, that means that potentially at the intermediate sites, we can target those sites for alternative strategies for drug development.

Alana Hippensteele: That's fascinating. What are some other treatment targets this AI can help to address through the development of therapies?

Dong Xu: Yeah, so one byproduct of methods that we can predict changes in amino acids in terms of the impact pretty accurately, so let's say you have a protein with amino acids, and then some amino acids might be changed, we call that mutation. It's actually not easy to predict the impact of mutation. But this method can predict for them for energy changes pretty accurately, much more accurate than our previous methods.

So with that, we first of all can study the impact of mutations in diseases, for example, in cancer, a lot of these cancers the proteins have mutations, some mutations play important role, some may not, and with this method, we can really see what changes can change the protein stability or other things. So that potentially can give us a drug target as to where we should target the protein. So it can provide some kind of insight, although it's not a direct drug development, it is a general method. So I have to say that this is not a drug development method, but it will help part of the drug development process.