Virtual Control Groups Could Reduce Animal Testing in Drug Development

Control groups are the backbone of rigor in scientific experiments. They give you a factual baseline you can use to compare against what you're measuring in your experiment. You can isolate the effects of a compound or drug with confidence that the biological response will be separated from variability or procedural artifacts.

Without an appropriate control, a hypothesis remains a hypothesis. It cannot be addressed, as it cannot be untethered from unreliable evidence.

That said, traditional control groups do carry limitations. They consume significant financial and biological resources and raise important ethical questions, particularly when conducting nonclinical experiments. In that type of experiment, biological complexity in humans is not always reflected in controls. Therefore, there is an inherent limitation between the controls and the real-world questions you're trying to address.

While these concepts, like digital twins and VCGs, may feel like a product of the current technological moment, serious consideration of tools like these gained traction in the early 2000s.

Today, we are at an inflection point in the maturation of computational power. The potential is palpable. We are starting to recognize the decades of accumulated knowledge, experience, and data that hold untapped scientific value.

This huge inflection point also converges with advances in machine learning, increased understanding of system biology, and high-dimensional biomarker profiling over the last decade. We now have a better understanding of which biomarkers are associated with which diseases and pathways. Once properly harnessed, all of these developments can provide us with credible tools, like VCGs.

However, in terms of biology, AI is still at the earliest stage of its journey. There are a lot of unknowns that remain to be researched, especially in different systems: in vitro, in vivo, or in silico. Initiatives are pushing toward model-informed drug development, with publications demonstrating that computationally derived control arms can produce comparable outcomes to traditional ones. This has shifted the conversation from theoretical to practical.

We're definitely not at the beginning of acknowledging the potential of computational approaches, but we're at the consequential chapter of it.

This question speaks to one of the most important ethical commitments of our field, the 3Rs: replacement, reduction, and refinement. There's a lot of talk about a fourth R: responsible or regulated. The Rs have long served as the moral compass in non-clinical research.

VCGs represent one of the most promising tools that I've seen so far that honors these principles in an evidence-based way. In practical terms, they significantly reduce the number of animals assigned to a study. This stands even if you are partially replacing a control arm.

VCGs can be implemented retrospectively or prospectively. For example, retrospectively, we have investigated how VCGs would have concorded with existing experimental conclusions. Prospectively, researchers want to predict study outcomes. This will likely require a stepwise approach. As you conduct an experiment, you assess the concordance level between nonclinical models and VCGs. Where are the unknowns that were captured? Did you uncover something new? Once these questions are addressed, you don't have to continue running traditional VCGs alongside, as long as you have confidence and are mitigating potential risks.

The value of doing both retrospective and prospective analysis is that you can test questions to predict outcomes, how you'll address them, and how you'll mitigate risks in a real-world experiment.

At the moment, VCGs are most powerful when they are used alongside in vivo data, adding a layer of computational validation to defend it, to be transparent about how the data was collected, and how it's used, so it strengthens your conclusions of a safety assessment study to produce an ethical and reliable report of that drug’s development.

There's a trust that needs building, which will require support, education, and collaboration. For the near future, the implementation of VCGs may look more like a hybrid model, achieving partial replacement and reduction.

VCGs will have limitations. Firstly, they will not be amenable to every single study design, and when they are, they will need to meet the same scientific rigor and validation standards. A study is going to change the patient's life. We don't want to put the patient's safety in any jeopardy; this is of utmost importance.

Currently, the most significant limitation of VCGs is the lack of data and the way it's standardized. Is the data of good quality? Is it fit for the purpose of being used as a replacement? It doesn't have to be a big database; it must be reliable. From definitions to laboratory conditions and endpoints, information needs to be reflected in data sets. What a study needs and what is available in databases will determine how best to get the VCG as close to equal as possible to the required conditions.

Another limitation—and I wouldn't call it just a limitation, more a journey—is uncertainty: what are the limitations that are not yet known?

There's probably going to be slow adoption at the beginning, but then we're going to see an uptick once confidence, trust, and data grow. Bringing together regulators and industry consortia to establish standards will help accelerate adoption and set the criteria for data. While the Food and Drug Administration and European Medicines Agency, have signaled openness in terms of new approach methodologies, they're not yet globally harmonized. They have not issued a framework or guidance yet.