Accelerating biomedical innovation with AI through collaborative iteration

How the Wyss Institute is using AI to transform biology by connecting computational insights with human-relevant experimental discovery and translation

By Mariel Schoen

Long before ChatGPT brought Artificial Intelligence (AI) into the public spotlight, scientists at the Wyss Institute were using AI-driven computational approaches, including machine learning and deep learning, to solve complex biomedical problems and develop new technology innovations.

Today, while much of the field is racing to build larger models, the Institute is focused on a different challenge: translating the growing capabilities of AI into better biological understanding and real-world impact. Computational approaches can provide valuable predictions and interpretations, but researchers need to combine them with experimental and clinical evidence to bring about meaningful advances. At the Wyss, these insights are tested and refined using human-relevant experimental platforms, multidisciplinary expertise, and an established translational infrastructure designed to move promising discoveries toward real-world application.

The promise of AI in biology goes beyond better predictive models and faster drug discovery; it lies in fundamentally improving how scientific discovery happens to accelerate patient impact and advance human health. This was the central message of the Wyss Institute’s recent Application-Driven AI Symposium, which brought together leaders in academia, biotechnology, pharmaceutical companies, Wyss startups, and technology infrastructure companies, including Amazon Web Services (AWS), Microsoft, Recursion, Eli Lilly, Novartis, MIT, and Tufts University, to explore how AI is transforming translational science and therapeutics development. Throughout the day, one idea surfaced repeatedly: we cannot look to AI alone to solve biology’s biggest challenges; we need to learn how to collaborate with it.

AI just shouldn’t predict things without validation. Its predictions should drive experiments, those experiments should refine the models, and that cycle should continue at scale.

As Emilia Javorsky, M.D., M.P.H., Wyss Institute Mentor and Director of the Futures Program at The Future of Life Institute, framed it at the outset, “AI just shouldn’t predict things without validation. Its predictions should drive experiments, those experiments should refine the models, and that cycle should continue at scale.”

For the Wyss, this philosophy has already been validated in practice. The Institute has helped launch multiple AI-enabled companies including Dyno Therapeutics, which applies machine learning to engineer more effective adenoviral vectors for gene therapy delivery; ReadCoor, which used AI for spatial tissue profiling and was later acquired by 10x Genomics; Ultivue, which also applied AI to spatial tissue profiling and later merged with Vizgen; Unravel Biosciences, which uses AI to compute gene networks to identify precision therapies for individual patients with rare diseases; and Manifold Bio, which combines machine learning with high-throughput protein engineering to create tissue-targeted biologic therapies. These companies demonstrate that integrating AI with experimental biology can produce real translational outcomes, validating both the science and the models themselves.

AI as a collaborator: connecting data, discovery, and people

Much of the conversation around AI focused on increasingly powerful models. But throughout the symposium, speakers argued that the real breakthrough isn’t larger models; it’s creating systems that connect people, data, experiments, and expertise.

Panelists repeatedly described “closing the loop” between AI predictions and laboratory validation as an organizational challenge rather than a technical one. “A lot of organizations seem to be running what looks like a loop, but it’s often just a relay race…there’s a lot of manual handoffs, a lot of waiting, a lot of information that gets lost in translation,” said Amrita Sarkar, Ph.D., M.B.A., who is the Global Head of HCLS GenAI Startup GTM at AWS. Creating a true feedback loop depends on generating high-quality data as well as building workflows that allow computational predictions, experimental findings, and scientific expertise to inform one another. Each iteration of the loop provides the opportunity to reduce uncertainty, sharpen the underlying hypothesis, and guide a better-informed next decision.

At the Wyss, treating AI as a collaborator means using computational models as active contributors to the scientific process, but not as systems that provide a single definitive answer. A structure or co-folding model might suggest how molecules interact, a perturbation model might predict how a biological system will respond to an intervention, and machine-learning approaches might identify patterns across imaging, multi-omics, chemical, or clinical datasets. Researchers evaluate these complementary insights by obtaining new experimental evidence to refine the AI models, build stronger hypotheses, identify uncertainty, prioritize promising directions, and determine what to test next. Scientists remain responsible for asking the right questions and validating the results. As Hananel Hazan, Ph.D. put it, AI is “very brilliant but overconfident.” Hazan is a specialist in bioinspired machine learning and neurocomputation at Tufts University’s Allen Discovery Center.

Building an ecosystem, not just technology

That collaborative philosophy is already taking shape through the Wyss Institute’s growing Translational AI Catalyst, which connects researchers with computational expertise, models, infrastructure, and external collaborators needed to integrate AI into translational programs. This year’s call for Wyss Validation Projects encouraged researchers to incorporate AI into translational projects, which resulted in 25 proposals strongly relying on AI (a 47% increase over last year), emphasizing how rapidly these tools are becoming embedded across Wyss research.

The symposium reflected the Wyss’ transdisciplinary approach to innovation. By bringing together academic researchers, staff with deep industrial experience, clinicians, technology companies, pharmaceutical leaders, entrepreneurs, and its own startups, the Institute is creating the collaborative ecosystem required to translate AI from prediction into patient impact.

Industry leaders discussed the infrastructure required to connect biological foundation models with laboratory workflows. Scientific leaders from pharma described how AI is transforming therapeutic design and accelerating drug discovery only to be held back by the same suboptimal animal models that have hindered pharmaceutical success for more than 80 years. Wyss researchers demonstrated how AI is being integrated with data from human-relevant experimental systems, including patient-derived Organ Chips and organoids. These platforms allow researchers to test computational predictions in human-relevant tissue and organ contexts using clinically relevant drug administration regimens, generate new evidence where existing data are limited, and refine their understanding of disease mechanisms or therapeutic responses. In some cases, the resulting data improve the model predictions and help the researchers to decide which hypothesis, candidate, or technology to focus on.

Rather than existing as separate conversations, these perspectives reinforced one another. Together, they illustrated a simple but powerful idea: no single laboratory, company, or AI model alone will transform biology. Breakthroughs emerge when diverse expertise converge around shared scientific challenges that are too difficult for any one collaborator, even AI, to solve on their own.

Translation begins with better biology

If collaboration is the engine, high-quality biological data is its fuel. Throughout the symposium, speakers argued that biological data will ultimately determine who succeeds in AI-enabled biomedical research and development. The quality of experimental systems, the questions scientists ask, and the rigor of laboratory validation become competitive advantages. As Wyss Founding Director Don Ingber, M.D., Ph.D., put it, “I’m less interested in the speed of the loop than the quality of the loop.”

That philosophy has long defined the Wyss approach. Across the Institute, technologies spanning genomics, synthetic biology, organ chips, organoids, protein engineering, imaging, and gene delivery generate rich, human-relevant datasets. Combined with clinical data from collaborators, these proprietary datasets ground AI in human biology, enabling predictive models that go beyond what can be learned from publicly available data alone. Each cycle of prediction, experimentation, and validation can generate smaller, information-rich datasets designed to resolve specific biological uncertainties, sharpen subsequent models, and guide the next decision, thus demonstrating that strategically generated data can be more valuable than scale alone.

AI helps researchers form new hypotheses, design better experiments, interpret increasingly complex datasets, and uncover therapeutic opportunities that might otherwise remain hidden. As Ingber reminded attendees, “The power is in human creativity and the collective that can use AI as a collaborator, as a deeply engrained tool.” At the Wyss, AI accelerates scientific discovery, but it does not replace it.

The power is in human creativity and the collective that can use AI as a collaborator, as a deeply engrained tool.

While AI has generated enormous excitement across drug discovery, many organizations are still working to translate computational predictions into real-world therapies. Nearly two decades of experience developing breakthrough technologies, launching startups, licensing innovations, and advancing therapies toward patients provide the translational infrastructure needed to move AI-enabled discoveries beyond prediction and into practice at the Wyss.

From insight to impact

Rather than asking, “How can we use AI?” the Wyss Institute begins with a different question: “What solution will solve our problem?”

As technology companies invest billions in increasingly powerful AI models, the greatest opportunities will belong to organizations that can pair those models with high-quality biological data, rigorous experimental validation, and a proven path to translation. The Wyss has spent years building this exact ecosystem: one where AI, human expertise, and translational science work together to move discoveries from prediction to patient impact. It’s also where new partnerships can accelerate the next generation of biomedical breakthroughs.

The real opportunity is in connecting insights from computational models with experimental and clinical evidence so that we can build a more complete picture, decide what to do next, and ultimately create real-world impact on patients.

Wyss Computational Biologist and co-leader of the Translational AI Catalyst Megan Sperry, Ph.D., describes this evolution as a shift from generating insight to enabling action. “The real opportunity is in connecting insights from computational models with experimental and clinical evidence so that we can build a more complete picture, decide what to do next, and ultimately create real-world impact on patients,” she said.

The greatest life science breakthroughs will come not from algorithms alone, but from organizations that combine advanced computation with high-quality, causal biological data with human relevance, rigorous experimental validation, and scientists who can collaborate across disciplines and the human-AI interface to ask the right questions that improve patients’ lives.

• Application-Driven AI Symposium Highlights