Drug safety and personalized medicine: a vital connection

With a PhD in organic chemistry and years of working with researchers in drug discovery to develop Elsevier’s data and information tools, Dr Catherine Noban opens in new tab/window is now applying her know-how to advanced AI.

As Lead Product Manager for Life Sciences Biomedical Innovation at Elsevier, she’s exploring innovative ways to better predict adverse drug reactions.

“It’s very motivating to work on a project related to patient safety,” Catherine says. “And it’s a natural fit since Elsevier also helps organizations accelerate the development of drugs to treat diseases. We are always exploring how we can support our customers in their innovation efforts to better predict adverse drug reactions using various Elsevier solutions — while also streamlining their R&D and reducing the need for animal testing.”

The work sits at the intersection of patient safety and tailored therapy — two areas she’s found to be intrinsically related. And she says it’s essential to understand this connection to get the most out of AI.

“Patient safety and tailored therapy are really two sides of the same coin,” she says. “But for now, we need to stay focused. It’s important to really understand the specific problem you want to address when you apply advanced AI techniques.”

How to make drug discovery more clinical

Elsevier Life Science Solutions opens in new tab/window supports various phases of drug discovery and development. For instance, customers may choose to search the drug approval documents and extracted data of PharmaPendium opens in new tab/window, visualize biological pathways via EmBiology opens in new tab/window, or go deeper into the bioactivity or molecular level via Reaxys opens in new tab/window.

“One of the key challenges of preclinical development is predicting how and why a certain adverse event will play out in a specific individual,” Catherine explains. “For example, what does this particular compound do to a dog? What does it do to a human — or different humans? When does the information correlate? When does it not correlate? So you try to understand the why using as much historical data as possible. And then you can draw your conclusion, make a hypothesis and test it.

This not only helps preclinical development teams with their hypothesis generation, she adds; it ultimately helps them improve healthcare outcomes in an ever-evolving healthcare landscape:

Basically, we want to help scientists understand the effect of a certain compound. We do this by helping them compare the new molecules they are making with ones we may already know a lot about. In other words, it’s about putting safety in the discovery stage — and thereby, down the road, decreasing the need for animal testing while increasing the chances of a clinical trial’s success.

Safety first in drug development

The importance of drug safety in drug development is clear. Drug safety issues are also one of the major causes of late-stage drug failure opens in new tab/window in pharmaceutical development.

It’s also clear that AI and machine learning are ideally suited to assess the possibility of connecting any relational dots of a drug's effects after it enters the market — and the even more dizzying implications of when this drug interacts, perhaps adversely, with those drugs already on the market. It’s simply impossible for mere mortals to cross-test the thousands of medicines currently on the market.

It’s also become clear that animal testing is often ineffective, and AI offers tremendous opportunities to contribute to the efforts related to the 3R principles: Replacement, Reduction and Refinement opens in new tab/window of animal testing. “I really hope our project contributes to this quickly changing landscape of assessing whether a medicine is safe or not,” Catherine says. She’s confident the project will be a success and hopes to have a proof-of-concept soon. “There are immense challenges, but we have the required expertise to work through them,” she says. “So it’s just a matter of time.”

Staying focused: one use case at a time

With a BSc/MSc in Biology/Biochemistry from Louis Pasteur University and a PhD in organic chemistry from Imperial College London, Catherine arrived at Elsevier in 2006. Hired as a subject matter expert to evaluate the quality of databases, she was soon getting more hands-on via later jobs in sales and product development.

“I always wanted to understand everything, so naturally, I loved science. Both parents are medical doctors, so that may have also been an influence,” she recalls and smiles. “But I didn’t love the lab work. And Elsevier offered the opportunity to still stay busy with the cutting-edge of science — staying up to date as everything evolves.

“I can talk to customers — people who really know their stuff — and delve into understanding what they do on a scientific level. Usually, their goal is to develop new drugs, so we really try to support them in accelerating the set-up and discovery process. In this way, they can focus more on their expertise: designing and doing the actual lab work and then the related analysis to take the program forward.”

Reaxys Target and Bioactivity “University”

Catherine got her “proper data schooling” while helping to develop what’s now known as the Reaxys opens in new tab/window Target and Bioactivity (RTB) platform, which was originally formulated to facilitate better connections between biological and chemical data.

“This was really the school I graduated from when it comes to knowing the data industry,” Catherine says. “I really learned it from scratch: how to produce such a massive and complex data set and what comes with it. It’s not enough to understand the data; you need to understand how the customers can best use it. You really need to understand both to make it work.”

More recently, she led the development of a prediction tool for safety margin risk assessment. A collaboration with a large pharmaceutical company, the project was centered around using different Elsevier datasets and systematically applying scientific rules to data to support very early safety pharmacology assessment. Now, the idea is to take these efforts to a whole new level.

In search of the bigger (molecular) picture

“As we explore how to support our customers in their preclinical assessments and continuous scientific monitoring,” Catherine says, “it’s important to investigate the possibility of looking at things more holistically using new technological advances — bringing together different elements to create a bigger picture. In this way, we can help scientists to better deduce and prove correlations, compare substances and properties, and do predictions. “But first, you must still understand the specific problem you want to address,” she adds. “As technology and science advance, there is a greater need to deep-dive into the actual problems. This is really key to data integration and applications of advanced AI techniques. Only then can you look for the relevant Life Science datasets and how to combine them to address your particular question.”

Built for scientists by scientists

“We’re dealing with scientists, so from the beginning, we focus on three things: quality, comprehensiveness and transparency,” Catherine says. “You need data you can trust and enough of it to provide context — the big picture — so you can put A and B together to see how they are the same and how they are different. Of course, it will never be perfect, and you also need to consider that.

“You also need to be transparent. We’re dealing with drug safety so you must have appropriate processes in place: where the data comes from, how it was cleaned, what any algorithms are doing.”

Everything is connected

“Our datasets are already pretty much set up like this: Here’s the structure, here’s the properties of the compounds, this is what happens to a dog, this is what happens to a human, etcetera.”

And indeed, discovering how individuals react differently to drugs or a combination of drugs is intrinsically linked to the rising field of personalized medicine, also known as tailored therapy.

“Just imagine the potential once we can better track what happens when a drug is approved and becomes available to the population. At this point, it becomes interesting for personalized medicine. This drug may now reach people who were not represented in the original trials. You can then study and analyze how it affects different populations.”

“Or perhaps the drug was even discontinued or withdrawn. Now this is very important: When it comes to personalized medicine, the failures can teach as much as the successes. And it’s very important to capture this information. And of course the ideal time to know this information is when you’re still in your discovery phase.” However, before the circle between discovery and clinical testing is complete, there’s a lot of work to be done. “This project may very well be more generally applied to other use cases,” Catherine says. “But first, we’ll stay focused on predicting adverse drug events.”

Learn more about how Elsevier datasets and tools are driving drug discovery and development opens in new tab/window.