The Future Is Ancient: Evolution, Metabolics, And AI

A conversation with Hannah Gordon, Ph.D., chief scientific officer, Enveda

Nearly half of all medicines taken in pill form originate from a molecule found in nature. But almost the entirety of the natural world remains a chemical mystery to scientists due to the time and enormous cost of purifying individual molecules from highly complex mixtures, identifying their structure, and characterizing their activity.

In this Q&A, Life Science Connect’s Morgan Kohler caught up with Hannah Gordon, Ph.D., chief scientific officer of Enveda, to discuss the next phase of AI-driven drug discovery and why the future may depend less on synthetic chemistry libraries and more on decoding nature’s evolutionarily selected chemistry.

Why are natural products and evolutionarily selected chemistry reemerging as a major frontier in biotech?

There's always been a lot of attention on drug discovery's failure rates, and these have barely budged in decades. As an industry, we're constantly trying to do better. One of the most proven answers is also the oldest: nature. About a third of all small-molecule drugs trace directly back to natural products — more than half if you count those inspired by them. Add in the fact that nature's chemistry also survives trials better and it's hard not to wonder “how is everyone not in this space?" The answer is both philosophical and technical, and I think both are driving its revival.

First, everyone mostly starts with biology: what protein do we need to hit to treat disease X? But that requires understanding the biological underpinnings of a disease. What about the diseases that don't have a clear, validated target? We flip this philosophical approach on its head and ask: where has the chemistry already shown us it's safe and efficacious? We start there. Then, using the most advanced biology and chemistry tools, we let nature show us how she did it. Not many chemical library starting points give you the fundamental confidence you get from natural products — it's the best place to start.

Second, the question has never been whether we believe in the potential of nature's chemistry. It's actually been a technical one. You can’t really control the concentration or complexity of nature’s compounds. In the 1990s, as synthetic libraries exploded, you could screen millions of made-to-order compounds fast and cheap. Next to that, natural products looked slow and messy — trace amounts, hard to isolate, hard to identify. A tougher path in an already brutal industry. None of those things are true anymore. The technology has changed: we can now read the chemistry of a biological sample at scale and use machine learning to find the molecules worth chasing. The old reason to skip nature's chemistry is gone, and that's why it's coming back.

In what ways are AI and metabolomics changing small molecule discovery?

Simply put, it's completely exploding our ability to learn from this chemical space. Historically, characterizing even a single natural product could take years, and doing it for millions of molecules at once was impossible.

This is what the partnership between metabolomics and AI changes. Metabolomics let us profile millions of molecular signals across diverse biological systems all at once. That's a lot of data — and the perfect problem for AI. Machine learning models can find the patterns, predict molecular structures, and help us prioritize the most promising compounds. This moves us from a largely empirical process to a predictive one at massive scale.

It's also worth saying just how much bigger that space is. Natural products occupy a fundamentally more expansive and diverse chemical universe than synthetic libraries, even when we filter for drug-like molecules. It’s crazy to think about, because another unlock is that our approach could change what we even consider "drug like." Almost every classic drug-likeness rule — Lipinski, Veber, and others — were built on synthetic chemical libraries. In fact, many are known to penalize natural products. Nature breaks those rules all the time. But we know from natural products' success rates that the rule breakers work. There's a successful equation to be learned there, and we finally have the data to learn it.

All of this means we can search far more chemical space, find molecules traditional methods would have missed entirely, and make better decisions earlier before we've spent time and money on the wrong ones.

How is Enveda using metabolomics and machine learning to improve hit identification and mechanistic understanding during the earliest stages of discovery?

For any molecule in a complex mixture, you really want to know two things: what is it (specifically, its chemical structure) and what does it do? This is exactly what our platform answers, essentially at the same scale as traditional high-throughput screens of a pure-compound library. We have built-in data-feedback loops so every time we see a novel molecule, we validate the structure and feed it back to our machine learning models. With every iteration, they get better. This has been tremendously successful thus far and has contributed to our deep pipeline of development candidates.

Outside of hit finding, we also want to understand the biology as early as possible. Our first job isn't to find the most hits; it's to set each asset up to succeed in the clinic. By integrating chemical, biological, and translational data from the very beginning, we start forming hypotheses about mechanism of action, target engagement, and disease relevance while a program is still young. That means understanding the targets and pathways well enough to measure whether we're engaging them and knowing what safety signals to watch for, if any, before we've committed years and capital. Front loading that understanding is how you make better decisions earlier, cut downstream risk, and improve the odds that the molecules you advance are the ones that can genuinely help patients.

Why is translational medicine becoming the critical bottleneck in AI biotech?

There's a difficult hurdle in drug development: what works in the lab often doesn't work in people. It’s a hard problem! People are diverse and complex even before we layer in disease. And there is a natural conflict between the complexity needed to keep a model accurate and the ease of screening it. I spent my academic career working across in vivo model organisms, using them to interrogate this question. The hard lesson is that we just don’t know what we don’t know until we know it. The ideal partnership is pairing translational models with methods to allow you to uncover what you don’t know. AI has made the front end faster than ever (e.g., generating hypotheses, identifying targets, discovering molecules). But that just moves the bottleneck downstream. The hard part now is knowing whether that insight will actually translate to help people.

Across the industry, we're generating more potential drug candidates than ever before. But many of these candidates are stacked on the same “validated” pathways, often offering limited improvement to patients. For novel first-in-class biology, the limiting factor is increasingly our ability to predict human biology and understand which programs have the greatest likelihood of clinical success. Our models are only as good as the data we put into them, and each model will have a bias (aspects of the disease it models better than others).

That's why translational medicine has become so important. It sits at the intersection of discovery science, biomarker development, clinical strategy, and human biology. Every drug program needs a strong “reason to believe,” and the translational model often serves this. At Enveda, we have an added advantage: nature made these molecules, and humanity has used them as therapies for generations. Highly translational models then help us home in on the best indication for each medicine. The companies that succeed will be those that anchor their reason to believe in the most trustworthy sources. I can’t imagine a more reliable one than nature.

Tell me about the shift from target-based discovery toward target-agnostic platform approaches.

For decades, drug discovery has started with a predefined target. Scientists identify a protein they believe is involved in disease and then search for molecules capable of modulating it.

That approach has produced important medicines, but it can also limit discovery because it depends on our existing understanding of biology. Many diseases are highly complex, and some of the most important biological mechanisms remain poorly understood. In fact, I don’t think it’s a coincidence that some of the largest areas of unmet need are unmet precisely because we don’t have a clear target for them. So how do we expect to tackle those diseases with the traditional target-first paradigm?

Target-agnostic approaches flip this. Rather than beginning with a target, we begin with biology. We look for small molecules that produce a meaningful biological effect and then work to understand the mechanism driving it. In many ways, it's a simpler framing, and one that more closely resembles the end symptom(s) we are trying to treat in a patient. Instead of trying to hit TNFalpha, we are trying to turn down the inflammation coming from a major driver of it: the TNFalpha pathway. Advances in AI, high-dimensional biology, and large-scale experimental systems are making this approach increasingly feasible. These technologies allow us to uncover previously unknown pathways, identify novel mechanisms of action, and expand the universe of druggable biology.

I think this is one of the most exciting shifts in all of biotech because it allows us to go after the biology that target-first discovery has not been able to crack. And these methods can synergize. Deploying target-agnostic approaches gives us space to learn novel targets and pathways that we’d been missing (but that nature has been using all this time). Knowing those targets feeds back into how we understand and model the disease, which can lead to even better therapies.

About The Expert

Hannah Gordon, Ph.D., chief scientific officer, Enveda, brings extensive expertise in translational biology and strategic drug discovery. With over a decade in biotech, including at Recursion, Hannah has become a leader in leveraging cutting-edge technologies for drug discovery. She excels at transforming complex scientific concepts into actionable processes, fostering cross-functional collaboration, and driving progress in translating nature into medicine. Her blend of molecular biology expertise and commercial acumen makes her a key innovator, continually pushing boundaries in drug discovery and development. She holds a Ph.D. in human genetics and an HHMI-sponsored MS in clinical investigation from the University of Utah. When Hannah is not chasing her curiosities, she can be found adventuring in the Rockies with her partner, Tyson, and dog, Carbon.