Virtual control groups work to cut animal use in toxicity testing

Dr Matthew Clark(opens in new tab/window) chuckles when asked to explain his work.

"My job is always difficult to explain to people," he says. "But basically, I use analytics and AI to make research more efficient. And it’s actually a nice change with this Virtual Control Group project: everyone understands it within 15 seconds."

Specifically, he is building a repository of historical animal testing data. The idea: Why repeat a control group if the information is already out there? Meanwhile, he also wants to go after the root causes behind why some animal tests are bad at predicting if a drug will have an adverse effect on humans.

Meanwhile, Matthew has worked in a startling number of different fields: logistics, drug design software, data extraction, watchmaking … Wait, watchmaking?!

"I was cross-applying similar technologies across dramatically different areas of research,” he says. “And in fact, the underlying maths of watchmaking is similar to molecular dynamics — which makes it relate directly to drug development."

Another recurring area of study for Matthew relates to animal testing — which makes him the perfect guest for Elsevier’s series “Successful Alternatives To Animal Testing.”

Which animal tests actually predict human outcomes

Some of the research he is most proud of took place while he was still at Elsevier. As a collaboration with Bayer(opens in new tab/window), it was one of the first studies(opens in new tab/window) to measure how well animal tests predict human outcomes.

"We had decades of animal testing data gathered by Elsevier from the FDA(opens in new tab/window) and EMA(opens in new tab/window) in PharmaPendium. But no one had really looked in detail at the statistics on how well each animal test predicted what happened in people — which is actually the whole point behind these tests. So we did the statistics and found out which ones are the best and which ones are the worst at predicting human response."

The resulting paper(opens in new tab/window) won the Translational Safety Pharmacology Publication Award in 2019(opens in new tab/window). And now, with this compelling evidence in hand, he is working with different parties to change regulatory policy worldwide to reflect these discoveries.

"If a test doesn’t work, it has no value. So why require it?"

The world of Virtual Control Groups

Currently, he’s laying the groundwork for Virtual Control Groups (VCGs). “It’s a very simple idea: we already have lots of data on healthy control animals — at Charles Rivers alone we have hundreds of thousands of studies at our disposal,” he says. “So with this, we should be able to get a really good idea of what an experiment’s baseline values should be. So why measure them again for your new study? And I believe the results will actually be better because of the statistical power you get from using so much data.

"It's not complicated, right? It makes you wonder why we didn’t start earlier."

Building a compelling case for change

But yes, now it’s time for the grunt work: "The science is straightforward, and now the real challenge is the same as with all data science: assembling the historical data, carrying it out and doing the analysis. We need to make that compelling case for the regulators. Regulators and the safety pharmacology community will have many questions to address before it becomes commonplace. So we need to find out what works and what are the limitations. And even if we can’t replace all animal control groups, I believe we can still reduce them substantially — which will already make a huge difference."

Within five years, Matthew hopes VCGs will be part of a larger toolbox that works to minimize animal testing across the board. "Everyone was surprised to discover that animal tests are not always as predictive as we thought," he says. "Now we have to find and demonstrate alternative methods that better predict human outcomes."

Diving into root causes

Matthew is already musing on future research projects — such as finding the root causes behind why some animal testing is unsuccessful: "We really don’t yet fully understand the underlying biological differences between, say, a dog and a human that explains why tests are not working to predict human toxicity. And the more we know about this, the more we can reduce animal testing and make it more efficient at predicting human outcomes.

"The first step for this would be to identify what data we need so we can start gathering evidence. Sure, Elsevier might have a lot of data available in its publications and various other opportunities that could be explored."

Onboard for innovation

In other words, collaboration will be key for future innovations. "And this Virtual Control Group project is already a very good example of this: it’s long-term and requires a worldwide agreement across the safety community and regulatory bodies. If we are going to use VCGs for actual studies, we need everyone on board. And I’m very optimistic we’re going in the right direction."

"It’s kind of weird: here you have a substance in a jar that you give to a human, and we’re not actually in a position yet to predict what’s actually going to happen."

Let’s get to work.