Artificial intelligence seems to be overturning every part of life. How about this one: AI and its country cousin, machine learning, working together to develop new drugs. To see how AI can help and what some of the risks might be, Federal Drive with Tom Temin spoke with the Food and Drug Administration’s Associate Director for Policy Analysis, within the Center for Drug Evaluation and Research, Dr. Tala Fakhouri.

Interview Transcript: 

Tom Temin I imagine this is something of great interest to CDER, where you work in FDA and FDA writ large. I imagine the drug companies, the manufacturers and developers they’ve got to be looking  AI. Fair to say?

Tala Fakhouri That is fair to say, in fact, we’ve received over 175 submissions for drug approval that included the use of AI and machine learning and drug development. And the use has really traversed the spectrum of drug development from drug discovery all the way to clinical research, to manufacturing and to post-market safety surveillance.

        Learn how DLA, GSA’s Federal Acquisition Service and the State Department are modernizing their contract and acquisition processes to make procurement an all-around better experience for everyone involved.

Tom Temin Yeah, that was really my next question. Where in the lifecycle of a drug does all this apply? Because at the development stage it’s really they develop new molecules, essentially. How could AI help in that stage? Let’s concentrate there for a moment.

Tala Fakhouri Right. So artificial intelligence and machine learning, for example, can be used to predict how specific proteins will fold or to predict certain targets for molecules that are already on the market or to discover new uses for existing molecules. This is something that we call drug repurposing. These uses are very exciting. And we think they may contribute to the development of safer drugs faster. However, a lot of the application of AI in that early phase of drug discovery is outside of what FDA regulates. But we still see submissions that will include information about the use of AI in that early stage of drug development.

Tom Temin And do the same worries apply for AI developed as applied to everywhere else? And that is, did they use sufficient and correct data such that the output is reliable? Will that molecule will do what they hope it will do. Is that the case? I mean, you worry about the data and the algorithms.

Tala Fakhouri One way that we evaluate the use of AI and machine learning in drug development, let’s say we got an application with AI being used in clinical research to predict outcomes for patients, predict how they’ll respond to a treatment. For example, the way that we would review this application would take into consideration the benefits and the risk of using this technology. Specifically, we emphasize the ethical use of AI. We emphasize issues related to transparency. We need to know, for example, the data that was used to develop these models. Is that data high quality data? Does it address issues related to bias, which may then lead to bias in the algorithm itself? We also look at the model’s performance to make sure that it is predicting or it’s performing in a way that is consistent with how the sponsor or the specific researcher had intended it to do.

Tom Temin Got it. And let’s move on to the topic of how AI could apply to the clinical testing, because that’s in some ways one of the longest parts of drug development. You might be able to come up with the new drug in six months, but then you’ve got to spend five years testing it. And that could be really controversial, I imagine, because tests take as long as they take and developments of after effects or cures take as long as they take. Can I speed that up in a way that you can rely on it? Am I asking the right question?

Tala Fakhouri You are asking the right question. AI can be used in clinical research. In fact, for us, on the FDA side, the majority of AI uses in drug submissions are in the clinical research part of the spectrum. AI can be used for outcome prediction. This is one of the strengths of AI and machine learning is its predictive power, so it can take information about the patients, for example, about their lab values, their demographics, and predict how they would respond to a specific drug and if they will respond to a specific dose. This is wonderful, because you could do things like dose optimization using this technology and it’s pretty fast. So we do expect it to expedite certain aspects of clinical research. We also know that AI, for example, is used for something that would be called patient selection and stratification. Finding the patients that would be able to respond to the drug is very important, AI can be used to be able to do that. There’s also new applications of AI that are very interesting to the FDA. For example, the creation of something known as a digital twin. So, for example, you would have a single arm trial. Where everyone is taking the treatment and then you would simulate what would happen to the specific patient had they not taken the placebo. So this is another application that we expect to see.

Tom Temin We’re speaking with Dr. Tala Fakhouri. She is associate director for policy analysis at the Center for Drug Evaluation and Research in the Food and Drug Administration. And for the FDA, what is it that you need in the FDA to be able to keep up with this? You hinted earlier that there might be an extension of regulatory oversight that you would need. And would that have to come from Congress, for example.

Tala Fakhouri For us on the FDA side, specifically procedure, the evidentiary standards needed to support drug approval remain the same regardless of the technology that you’re using. It’s very important to emphasize that, that the paradigm that we currently use has not changed. We are actively monitoring advances in AI machine learning, and we continuously engage with experts, whether through expert workshops or recently in May, we published two papers, two discussion papers, one focusing on the entire drug development landscape, and the other one more specifically targeting the use of AI in drug manufacturing. In that document, in both documents, we raise questions to help engage with the community, with stakeholders, and we hope to receive a lot of good feedback. The purpose of this discussion papers is really to be able to understand if there are areas or gaps where additional regulatory clarity is needed. But I can tell you as of now, with the 175 submissions that we’ve received, our evidentiary standards are the same. The paradigm that we’re using has not changed, because there isn’t a need to provide additional clarity as of now.

        Read more: Artificial Intelligence

Tom Temin But it sounds like you have the potential maybe for some additional rulemaking based on what these submissions say and where those gaps might be.

Tala Fakhouri So after we receive the comments on the docket for the two discussion papers, we plan to carefully and thoughtfully analyze all of the feedback that we’ve received. We plan to conduct public workshops next year to be able to address needs for the community, in terms of additional regulatory clarity. And if there is a need to provide future guidance, of course we’ll be happy to do that, because we want to make sure that this technology is used in a responsible way and used to develop new, safe, effective medications for the public.

Tom Temin And what about the requirements that FDA would have in terms of your own people and their knowledge to keep up with developments in AI and algorithms and how this is all being used? Because there’s many forms of AI, many sources of AI, and they’ve got to keep up with that no less than the drug industry.

Tala Fakhouri Right. So internally within the FDA, within CDER, we are conducting a lot of work internally to be able to bolster our workforce, make sure that folks are trained in the use of these technologies. You can take classes, you can attend seminars, but also in terms of hiring, hiring experts, that could help us better understand the use of this technology in practice.

Tom Temin A final question with AI do you anticipate just from your general sense of what’s going on in the world, that this has the potential to lower the cost of drug development and deployment? I mean, the ideal world, the latest cancer drug would cost as much as an aspirin or as little as an aspirin, probably that’s unlikely. Could this drive cost out of the entire lifecycle here? Do you think.

Tala Fakhouri Costing of drugs is outside of the domain of what I work on within the agency. But one can expect that if you have drugs being developed faster, this may reduce costs on all ends.

Copyright © 2024 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.