How techbio is shaking up pharma

All tech companies start from the same premise: “Our industry is big and bloated. The model doesn’t work, and we tech disruptors are here to change that.” Techbios are no different. We are united in our belief that technology should be able to change the pharma industry for the better.

New drug discovery is extremely inefficient. Estimates of the average R&D cost per new drug are less than $1 billion to more than $2 billion per drug, with clinical trials taking up to 10 years, and only around 10% of those drugs make it to market. And all the while, people are dying of diseases we should be able to treat.

Why? Because we don’t properly understand biology yet.

Today, when researchers search for new drugs, they’re fighting against the odds. In absence of the right AI to make sense of their mountains of data, pharma companies often end up relying on intuition and human expertise. Now, imagine a world where AI can sift through those mountains, from gene sequences to protein structures to clinical trials, to predict the properties of novel drugs, and design targeted therapies.

As a physician and techbio founder, I believe that the best data coupled with the best AI is the key to unlocking complex biology. That’s why I built Owkin differently. Instead of focusing on one vertical segment, we work across drug discovery, development, and diagnostics. Understanding the underlying biology can impact every step of the chain driving improvements across the full stack.

AI is only as good as the data it learns from

A big problem when studying human biology is that most research has to be done on animals or simple human cells (like HeLa cells). Drawing parallels between these animals or cells to human biology is dangerous: After all, you wouldn’t study Spanish and expect to understand Portuguese. To bridge this translational gap, you need to work from real patient data: clinical data from doctors’ evaluations, patients’ genetic data, tissue data from patient tumors, and so on.

But where will your data come from? Is it deep, clinical-grade data, matched to your scientific needs? Have you built trust to access said data? This is where techbio can do what pharma cannot—build partnerships with academic and clinical centers to access the right data.

At Owkin, our approach has been to spend years nurturing relationships and building trust by putting patient privacy first through federated learning (a way to train AI models without the data ever leaving the hospital). We’ve found that by partnering with institutions, offering our expertise in AI, and ensuring data is accurate, we’ve been able to partner with more than 65 top academic centers, including nine of the top 20 oncology and general hospitals.

There is also a growing realization that more kinds of data are going to be needed to unlock AI’s full potential. This is why we built MOSAIC—a dataset 100 times larger than any in existence—to use spatial omics to decode biology. This type of data allows us to see cancer in a way we never have before, as though we could see the orchestra for the first time after only hearing its music.

Discover new patterns and causality

So once you have the best data, what do you do with it? You set out to create the best AI possible.

The great thing about AI is that it has the potential to uncover patterns humans could never find. For example, Owkin developed an algorithm that can look at a scrap of tumor tissue and accurately predict which genes are being expressed and where.

These tools are the first wave of improvements on the old industry playbook. The beauty of being an end-to-end techbio is that advancements in one area of AI feed the entire chain. We can build on the same AI to pinpoint drugs for patients who share the same biology, design more efficient data-driven clinical trials, and build diagnostics to match patients with the right treatments.

But we haven’t even taken the biggest leap of all: causality. Understand causality and we can understand the real mechanisms of disease. This will require new kinds of data—data where conditions can be controlled and single elements of an experiment can be changed, to see if that leads to disease states. This kind of “perturbation” data will be hard to find in humans, that is one of our challenges. But that is not the greatest challenge.

Human intelligence alone won’t cut it

Copernicus needed his telescope to study the stars. Robert Hook needed the microscope to see the first cell. To study ourselves, we need a new way of understanding, a new kind of intelligence that augments our own. We need to build a scientific artificial general intelligence (AGI).

Imagine an intelligence made up of hundreds of specialized AI agents that could run millions virtual experiments and choose the best one to run in a real, automated lab, using patient-derived organoids (3D miniaturized versions of organs and tissues grown from patient cells). An intelligence that could then feed their results to our researchers and data scientists, combining the best of human and artificial intelligence to bring us closer to causality.

I believe several key technologies are now mature enough to start assembling an AGI to tackle the intricate challenges posed by biology, that human intelligence has failed to fully decipher. We are inventing the right language for agents and data scientists, with seamless access to biomedical literature, patient data, and foundation models. This will allow us to do with a few thousand people, what pharma does with 200,000.

This is a race to understand life. To improve care with precision and scale for every patient. For our loved ones. For us. Because nothing else really matters.

Thomas Clozel is cofounder and CEO of Owkin.