A seed-stage AI company teaching machines to model how a drug ripples through human biology - predicting efficacy, toxicity, and why trials fail, then explaining the mechanism behind every call.
Somewhere on the fourth floor of a building on Montgomery Street, a model is running a clinical trial that will never touch a patient. Compounds bind. Pathways light up. A cell line reacts, ropes of signaling cascade, and a number appears: this drug is likely toxic, and here is the mechanism that makes it so. No pipettes. No mice. No eighteen-month wait for the readout. This is what Syntensor does on an ordinary Tuesday - it asks biology a question and gets an answer back before lunch.
The pitch sounds like heresy to anyone who has spent a career in a wet lab: that the cell, that gloriously messy bag of chemistry, can be written down as equations and solved. Most of biology has quietly agreed it is too complicated to simulate. Syntensor was built on the opposite conviction - that biology is a dynamical system, and dynamical systems, however unruly, can be modeled.
The stakes are not academic. Roughly nine out of ten drugs that enter clinical trials fail. Each failure is a story that ends badly and expensively, often after years of work and hundreds of millions of dollars. Syntensor's bet is that many of those endings are legible early - if only you had a model faithful enough to read them.
The platform pairs dynamical systems simulation and neural differential equations with large-scale representation learning over multi-omics data. Feed it a compound and a cell line; it simulates the assay, predicts pathway activation, efficacy indicators, and toxicity endpoints, and - crucially - hands back a causal hypothesis rather than a shrug.
Models how chemical perturbations propagate through cell signaling pathways, standing in for costly preclinical assays.
Forecasts drug response, toxicity endpoints, and the drivers of clinical trial success or failure from cell-line and compound data.
Genomics foundation models for gene-expression and trait prediction, integrating multi-omics signals at scale.
Neurosymbolic, interpretable outputs that generate causal hypotheses and visualize a drug's mechanism of action.
"Building an accurate mechanistic model of human physiology - to make complex biological processes tractable at scale for scientists, drug developers, and clinicians."
Plenty of companies now promise AI for drug discovery. What separates Syntensor is a refusal that sounds almost old-fashioned: it will not ship a black box. In an industry where "the model said so" is not a sentence a regulator or a medicinal chemist will accept, interpretability is not a feature - it is the point. Syntensor's outputs are meant to be argued with, traced back through the pathway, and either confirmed or killed at the bench.
That design choice traces straight to the founders. Clayton Rabideau earned a PhD in Chemical Engineering and Biotechnology at the University of Cambridge, where computational synthetic biology gave him the strange idea that cells could be simulated in the first place. Rosie Higgins arrived from the other end of the telescope - a Contemporary Media Practice degree, then product leadership at BenevolentAI and Novartis, where she learned exactly how emerging biology tech does and doesn't reach the market.
PhD in Chemical Engineering & Biotechnology, University of Cambridge. His computational synthetic-biology research became the seed of Syntensor's core technology. Now describes his work simply as "building foundation models for cellular biology."
Former COO at BenevolentAI and VP at Novartis, with deep product experience bringing emerging biotech to market. Came to computational drug discovery via an unlikely start in Contemporary Media Practice.
Figures are drawn from public sources (Republic, Kingscrowd, Crunchbase) and are approximate where noted. Revenue and full cap-table detail are not publicly disclosed.
The customer is the drug developer - the pharma or biotech team staring at a shortlist of candidate compounds and a budget that only stretches to test a few. Syntensor's promise is to tell them where to look: which molecule is worth the assay, which is quietly toxic, which mechanism explains a surprising result. It doesn't aim to replace the lab so much as to point it.
It runs in a crowded, fast-moving neighborhood - Insilico Medicine, Recursion, Cradle, and the broader race to build "virtual cell" foundation models. Syntensor's wager is that mechanism and interpretability, not raw prediction alone, will be what an FDA-facing industry actually trusts.
Back on Montgomery Street, that simulated trial finishes its run. The number lands, the mechanism is drawn, and a chemist somewhere decides not to spend the next year chasing a compound the model just talked her out of. That is the whole ambition, stripped of the jargon: to move a few of biology's expensive endings earlier, into a place where they cost a computation instead of a career.
Whether Syntensor's virtual cell proves faithful enough is still an open question - eleven people are a small crew to model human physiology, and biology has humbled bigger teams. But the direction is unmistakable. The lab of the future may open with a simulation, and the trial may already know how it ends.