Field Notes: Outpost Bio

Field Notes is a new content series from Seedcamp, which takes a closer look at the founders we back and the problems they’re solving. 

First up, we’re heading into one of the most poorly understood (yet consequential) new frontiers in healthcare: the human microbiome.

We are only just beginning to understand the human microbiome.

Genomic sequencing cracked it open: by reading microbial DNA straight from a sample, researchers could finally take stock of the vast community of microbes living inside us.  We learned that each of us carries a microbial ecosystem as distinct as a fingerprint – and while our human DNA is nearly identical from one person to the next, our microbiomes can differ enormously. These tiny, yet highly consequential, organisms shape our digestion, immune systems, and how we respond to both medicine and food.

Now, a new wave of technologies is taking the field one step further. Sequencing told us which microbes are present. Now, advances like metabolomics, deeper microbiome science, and AI models trained on rich biological data are beginning to reveal what those microbes actually do — and how they might impact our health.

This is where Outpost Bio comes in. Founded by Jenny Yang, PhD (University of Oxford, Marie Curie Fellow) and Alex Merwin (AWS Health & Bio), Outpost are building the AI layer for human microbiology – with openness at its core. Jenny and Alex feel strongly that the human microbiome is too vast for any one company to map alone, so Outpost keeps an open foundation to its work, open-sourcing key models and publishing through peer review.

We caught up with Jenny and Alex to discuss why it’s the microbiome’s moment, and the importance of an open source ethos to scientific integrity.

Q: Could you give me the elevator pitch for Outpost Bio?

Alex: We’re building the AI layer for human microbiology. We decode the human microbiome and make it much easier for anybody producing new medicines, ingredients, or products that improve our health to get those products to market quickly and safely.

For example, many medications interact with the microbiome in unpredictable ways that can make them intolerable — an upset stomach, or a nasty skin rash. This could result in the patient refusing to take the medication, and never experiencing the benefits. Or the microbiome can even metabolise the active compound, so even though you take the medication, you never realise any therapeutic benefit from it.

Q: Was there a single lightbulb moment when you realised that this technology needed to exist?

Jenny: I’d spent years in personalised health and medicine, a good chunk of it looking at human DNA and human genomics. Conceptually, I thought the microbiome was a hugely important space to keep pushing personalised health forward.

We’ve seen how our DNA affects our health. We can identify the genetic mutations that lead to particular subtypes of disease, and people are now building targeted therapies for them. But for all the progress in precision medicine, we still haven’t really moved the needle on why two people with the same mutation, on the same medication, respond so differently. And it’s not just medicine — the same is true of how differently we each respond to the same food. One key missing piece is the microbiome: whether it’s a drug or what you eat, it passes through your microbiome, and the bacteria there break it down in different ways. We’re nearly identical to one another in our human DNA, but our gut microbiomes can differ enormously — so if you really want an appreciation of personalised health, you have to consider the microbiome.

Q: What role does AI play in this field?

Jenny: The microbiome is incredibly diverse, and its orders of magnitude larger than human genomics. There’s vastly more genetic diversity to account for in the microbiome than in human DNA, so you’re working with very large, high-dimensional datasets. It genuinely requires the new, more sophisticated algorithms and the cloud compute infrastructure that can now crunch through all this data. 

Q: Was there anything you set out to build differently with Outpost Bio versus what you’ve seen in other biotech startups?

Alex: For me, the key element is scientific and academic integrity. There’s a lot of snake oil in the microbiome space, and a lot of overinflated startup claims. Jenny is very calibrated about scientific reproducibility. 

Jenny: I’d found a lot of people were so enthusiastic about AI that they’d immediately turn every problem into a purely data-science problem, even in very novel areas. I kept asking: how can you do that with complex biology when there isn’t a body of evidence in the literature to train AI on? So when we started Outpost, we first created that body of evidence – the data pack – that would justify building models capable of making predictions. Your model is only as good as your data – garbage in, garbage out – and that holds especially true in a field as complex as biology. So we focus on reproducibility and validation: rather than generating data in-house and calling it the gold standard, we also generate it with external research organisations, so we can genuinely test reproducibility and validate our findings.

Q: Why did you decide to open source your model?

Jenny: The biggest reason is giving back knowledge to the community. Everyone knows the microbiome is important, but there’s still so much to discover. It’s a huge opportunity, and it’s too big for one company to tackle alone.

Putting forward an open-source model, or publishing through peer review, also holds us to a certain standard and acts as a second shield to make sure any data we put into the world is rigorous. A lot of companies keep everything proprietary and behind closed doors, and I think that’s dangerous for science. You’re effectively saying, “the science we do in-house is correct, just trust us.” That’s not how you earn the trust of the doctors, patients, and wider public consuming these products.

Alex: Everything Jenny is saying is also true strategically. We genuinely want to be a force for good, and we also need to build a business, but the order matters: we set out to build in the open, then find a way to build a business around it, not the other way round. 

From a go-to-market perspective, there’s a lot of activation energy in getting a large partner to adopt a computational model for microbiology when they’ve never done it before. It’s far lower activation energy for them to download something under Apache 2. So commercially, open source just works for us.

It’s also how we fulfill part of the Pledge 1% we signed, where we donate 1% of profit, 1% of employee time, 1% of founder equity, and 1% of our product to charity. And it’s great for recruiting, too, because we really mean what we say and that is attractive to candidates.

Q: Has there been anything that’s surprised you on this journey?

Jenny: Two things, and they’re connected. The first is on reproducibility. I knew going in that validation mattered, but I underestimated how much we’d learn from the act of doing it rigorously. Every time we reproduced a result with an external lab, we learned something about the method itself — where it was fragile, where it held up. That’s fed straight back into better tooling. We’ve been able to build infrastructure that makes reproducibility the default rather than an afterthought, and it lets us validate findings with far more confidence.

The second surprised me more. We expected the models to be fairly domain-specific — that what you learn from one kind of microbial community wouldn’t tell you much about another. What we’ve actually found is that the models pick up much more general patterns across very diverse communities than we anticipated, and that knowledge transfers — we can take what’s learned in one setting and apply it in another. That has real implications, and without giving too much away, we’ve got results coming soon that I’m genuinely excited about.

Q: What’s next for Outpost Bio?

Jenny: The near-term focus is delivering on the results I mentioned — getting them validated and out into the world the way we hold ourselves to: built into the open model so anyone can scrutinize and build on them. We’ll keep expanding the data pack too, with more communities and more external partners, because the breadth of what the model has seen is what makes those general patterns hold up. 

Alex: And on the commercial side, it’s about getting the model into the hands of the people building real products — the first partners deploying it in their own pipelines. Getting started is meant to be frictionless — download it, try it, see that it works for yourself. From there we grow with them. But the order we set out with hasn’t changed — build in the open, be a genuine force for good, and let the business grow out of that. If we get the science right and keep it trustworthy, the rest follows. The microbiome is too important (and too big!) for us to do it any other way!