Olio Labs' AI Claim Rests on Wet-Lab Neuroscience - and That's the Point

The press release says Olio Labs has released a study showcasing its "AI-driven in vivo platform" that can "predict clinical trial outcomes." If you read that and picture a machine learning model ingesting trial data and spitting out success probabilities, you probably have the wrong picture.

The actual study is a piece of classical neuroscience. Female mice have nearly double the GLP-1 receptor expression in brain regions that process nausea, compared to males. Efficacy and tolerability of GLP-1 drugs like semaglutide and tirzepatide vary with the estrous cycle phase. Women in real-world medical records experience 2.5 times the rate of persistent nausea and vomiting.

None of that is AI. All of it is hard-won wet-lab and epidemiological work - single-cell RNA sequencing, species-specific behavioral assays, and analysis of 120 million electronic medical records.

I want to sit with that mismatch because it's more interesting than it first appears. The "AI-driven platform" framing is a fundraising narrative. The neuroscience underneath is real. The question is which one is going to scale.

Olio Labs was founded in 2023 by David Tingley and Tom Roseberry, both systems neuroscientists who previously worked together on the computational biology team at Loyal for Dogs, where they used research tools to manipulate single neurons. They raised $4 million in seed funding through Y Combinator's S23 batch, then another $5 million in March 2025 from Overwater Ventures. Their total is $9.0M in total across 2 funding rounds - a small pile by AI-biotech standards. Hologen, a London-based AI clinical-de-risking company, is raising $150 million at an $850 million valuation. PhaseV pulled in $50 million in 2025.

Olio Labs is not playing that game. Or at least, not yet.

Their public-facing platform, the "Combination Design Engine" or CoDE, is described as a system that searches trillions of potential drug combinations using machine learning trained on protein interaction networks, neural circuit maps, and patient data. That sounds like what the press release means by "AI-driven." But the GLP-1 study - published on bioRxiv in March 2025 - leans much more heavily on the laboratory side. The study constructed a mouse single-cell transcriptomic atlas of brain and body regions relevant to GLP-1 action. They developed novel species-specific in vivo phenomic assays to quantify aversive behaviors. They correlated circulating estrogen levels in humans with nausea and vomiting risk.

The AI is there, in the sense that they used computational methods to integrate data across these layers. But the insight - that female mice have higher receptor expression in nausea-processing circuits, that the estrous cycle shifts both efficacy and tolerability, that cycle-phase dosing could reduce side effects - comes from the biology, not the algorithm.

That might sound like a criticism. I don't think it is.

The more interesting companies in AI-driven biotech right now are the ones doing the boring work first. The ones that build real biological ground truth so their models have something true to learn from. Most of the space is still trying to retrofit machine learning onto incomplete or noisy datasets. Olio Labs appears to be doing the opposite - building a dataset from the ground up that their future models can actually train on.

There's also a simpler story about why this matters. Women make up roughly 70% of GLP-1 patients. About 40% of people who have been treated with a GLP1R agonist and stopped cite these side effects as the reason. If the study's suggestion holds - that adjusting dosing around high-estrogen phases could mitigate the worst nausea and vomiting - it could meaningfully improve adherence for the majority of this drug class's users. You don't need an AI platform to see why that's valuable.

But I haven't seen evidence yet that Olio Labs has moved from this GLP-1 observational study to an actual drug candidate in development, let alone clinical trials. The company claims its lead combinations target obesity and "are more effective with fewer side effects than Ozempic" but that language hasn't appeared in a peer-reviewed paper or regulatory filing. I couldn't find a clinical trial enrollment, a partnership announcement beyond advisory board names, or a disclosed pipeline outside the GLP-1 study.

That's the data gap. And it's a big one.

A study showing sex differences in drug response is a credible piece of science from two people with the right background. A platform that consistently predicts clinical trial outcomes across disease areas is a different order of claim entirely. The jump from the former to the latter requires validation that I haven't found - ideally in a setting where the model or platform predicts an outcome that was later confirmed or refuted prospectively, not retrospectively fitted.

What I can say is that the people seem right for this kind of work. Systems neuroscientists who've done single-neuron manipulation are not the profile you expect from an AI pitch deck. Patricia Martin, former COO of Eli Lilly's diabetes division, sits on the advisory board. That kind of connection to the company that literally built the GLP-1 category suggests someone who's been inside the problem Olio Labs is trying to solve.

The way to evaluate Olio Labs right now is probably not to bet on the platform claim. It's to watch whether their next study does something equally specific and unexpected in a different disease area, and whether the "in vivo phenomic screening" they describe actually catches a failure mode that traditional preclinical models miss. If the platform is real, it will show up as a pattern of these kinds of studies - each one surprising, each one grounded in actual biology, each one making the next one easier to run.

The test is simple: does the third study come as fast as the second? Or does the company stay one paper deep, with a $9 million war chest and a press release doing the heavy lifting?