Q&A With Dr. Frank Yocca, Chief Scientific Officer of BioXcel Therapeutics

Dr. Frank Yocca
Chief scientific officer
BioXcel Therapeutics

Companies like BioXcel hope to show to pharma and life sciences industry how new technologies like AI can benefit the industry and should be embraced. Dr. Frank Yocca, chief scientific officer at BioXcel Therapeutics, spoke with Medical Device & Technology about how he used AI during the drug re-innovation process.

(MDT:) Can you explain the work BioXcel has been doing with AI and drug development?

Yocca: There’s two things that you need to be clear about in regard to BioXcel Therapeutics. First is that we are a re-innovation company where we take drugs that have been in Phase II, Phase III, and some have even made it to the marketplace, and we reinvent it using new strategies for new uses. There’s a lot to it, and while AI is at the foundation, the strategy that we built was built on the ideas of several individuals on how to be successful in an area like neuroscience where the probability of success when you start out is less then 5%.

Back in 2015, when I started working with a company owned by BioXcel, we had to come up with a strategy that would give us a chance to increase the probability of success. Around that time, many of the big pharma companies were getting out of neuroscience research because of the probability of success being so low.

We came up with a unique strategy to use AI to re-innovate molecules. We also had to determine what kind of molecules we would choose from the AI to develop, how to do so, and could we do something to increase the probability of success. We decided to start out focusing on Phase II drugs. It can take six-to-seven years and cost up to $150 million to get to Phase II. Instead, we said let’s find an area that we think has unmet need and focus the AI on it and hopefully it can point us in a direction where can innovate something and start out in Phase II.

(MDT:) Could you go into more detail about how the AI works?

Yocca: I’m sure you’ve heard this from a lot of people, but artificial intelligence is not a black box that will simply provide answers. What we do is deploy an armamentarium of AI tools to help answer specific questions, each addressing specific problems in the drug re-innovation process. We call this composite AI because it takes several different things into account as it does the evaluation.

Our AI helps in obtaining information about drugs, targets, and indications, which are all connected in label property graphs (or knowledge graphs). These are workplace for AI. This also includes machine and deep learning methodologies to learn drug-like behavior of compounds. Our compendium is made up of all the drugs that exist, so we must have a way to narrow these things down. We also use natural language processing to extract relevant information from text, as well as graph-based data science, including methods to detect hidden information in the knowledge graphs. That’s where you find the gold, so you must get into the knowledge graph and look at it from a three-dimensional space to see where unique connections from these targets and systems, particularly in neuroscience.

In oncology, you target a tumor which is invasive and causes problems, so you want to eradicate it. In neuroscience, especially in the area of psychiatry diseases, it’s a conglomeration of different systems that goes array. That’s why the most successful drugs have been dirty drugs, or drugs with multiple mechanisms. As you get more specific, the drugs tend not to work as well because they haven’t focused in on the correct systems yet.

We’re working on recommendation systems to reveal hidden information in these knowledge graphs, using techniques such as matrix fractionization. This is a similar technology that companies like Netflix use to predict what other movies you may like after it has a sample of what you’ve been watching.

Ultimately, we look to predict what drug can be re-innovated for which innovation. We use the AI the augment that decision process when it comes down to drug innovation. We use big data, both structured and unstructured and use AI to detect the signal amongst all the noise.

It’s a daunting task. Over a million papers are published every year in life sciences, which is more than two-per-minute. AI allows us to process all of that. In essence, when you talk about AI, what are you talking about? It’s about how you handle data, and how you handle it in a way where it gives you information back and connecting it somehow.

AI can only truly leverage data when it is appropriately contextualized by carefully refining the questions you’re asking it. You only get out of it what you put into it, and how you get the information out is critical. AI is really all about asking the right questions, so you get the right answers.

We’re also looking at ways to leverage these large language models, such as GPT, and we’ve implemented bidirectional encoder representations from transformers model specifically that’s trained on biomedical data, creating our own GPT. We’re deploying this composite AI across the value chain (discovery, development, regulatory, C&C, etc.) and all the things that you need to have information on to develop a new drug.

(MDT:) Is it accurate to say that the AI is being used mainly to sort and quantify the data while all of the analysis is being done by the doctors and researchers?

Yocca: In some extent, yes. What we’re looking for are the outliers. Particularly when we’re looking for a new molecule, we’re looking for something that’s unique. That’s why I talk about the three-dimensional space because you’re looking at circuits in neuroscience and you don’t know how they come together. With the information that the AI provides, you get connections in unexpected directions, and you start following it up and see where the AI was going. You then must make a decision about whether it makes sense and if it’s possible to affect the physiology of this system by going in that direction. That’s when the neuroscientists and data scientists decided to set up a study to test it.

In terms of other areas, we’re applying AI to clinical situations. When we have a drug that gets approved, the more patients that use the drug allow us to understand what those that are non-responders versus responders. We can see if we can predict which patients will respond to the drug as opposed to those that won’t.

When it comes to the commercial element, the question is where do we think the drug will get the most traction and why. One of the things we’re looking at is which hospitals have the most patients that come in with agitation. There’s also a lot of assaults that happen in hospitals. When a patient comes into a hospital setting and is demonstrating moderate agitation, the situation at the hospital may cause an escalation. They’re in a strange situation, people are firing questions at them and then somebody pulls out a syringe. The next thing you, you’re dealing with a hostile patient. Based on this data, we’ve learned that it may make sense to put patient on medication right away, along with doing a verbal de-escalation.

(MDT:) How did you use virtual reality for this process?

Yocca: The virtual reality is great because what we do is project three-dimensional interactions between all the different indications, and you can actually step into it. You can look around in the virtual reality to see exactly the depth and different connections. Whenever somebody visits us, we put them into the knowledge graph, and we explain to them that this is how the AI started picking up new directions for existing drugs. They get a big kick out of it because we’re unveiling ideas and directions that are blind to you in normal circumstances.

(MDT:) So is it accurate to say that you’re using AI to allow researchers to see things from new perspectives?

Yocca: Yes, absolutely. The things that we understand have most likely been plowed in terms of a direction to build a drug. We want those new directions, which we can only get from the AI.

(MDT:) What future applications with this technology are in the works right now?

Yocca: What I see happening is the GPT is taking everyone by storm. The one that open AI offers is more of a generalization, and you have to work on it to make it very specific to make it what you want. That’s where people are going with AI: building their own GPT so that it will help them in their business.

I understand the fears that people have with this, but it’s all about what you do with it. The AI doesn’t create things itself, especially things that could be harmful. You put controls on it to do the things that you want to do. The benefit of AI is that it helps you think in different directions. It puts you in situations that may be uncomfortable, but you may be about to do something interesting.

(MDT:) How much do the results from AI depend on the person using it?

Yocca: Correct. What we’ve done here is made sure that our AI team is made up neuroscientists and data scientists. The data scientists start learning the neuroscience, and vice versa. They blend in together and their goal is to find something new that will be better than what exists now. They’re focused in on that and by learning each other’s craft, they can speak to each other more and more on a daily basis on what to look for and how to do analysis. It’s a powerful thing when you have a data scientist that understands an area very or a neuroscientist that knows how to deal with data on a daily basis.