Outer Biosciences Trains AI Models on Skin Kept Alive for Weeks

Outer Biosciences, a four-year-old startup co-founded by Michael Polansky, has spent years keeping donated human skin functioning outside a human body for stretches approaching a month, then feeding what happens to that tissue into predictive AI models. The company detailed its process publicly for the first time in a profile published by TechCrunch, after years of near-total silence about what it was building. The bet: skip the animal proxies and lab-grown organoids that dominate dermatology research, and train models directly on how real human tissue responds.

Start with what the model actually does. Before a single chemical touches living tissue, Outer Biosciences runs it through a ranking system that scores unproven compounds by how likely each is to improve a chosen skin function, say pigmentation or barrier strength. The top candidates then get applied to the tissue, and the outcome loops back in to sharpen the next round of rankings. That closed loop, Polansky told TechCrunch, is what the company actually sells, not any single ingredient.

That loop only works because of how long the tissue survives. The skin comes from surgical waste, mostly plastic surgery, that would otherwise be discarded, and Outer Biosciences says it obtains the material through a screened network of nonprofit and for-profit tissue suppliers, principally the National Disease Research Interchange and the Cooperative Human Tissue Network, both of which draw NIH and National Cancer Institute funding while operating independently of the federal government. Polansky said the arrangements proceed under what the company describes as “institutional review board oversight and documented donor consent,” and that every sample reaches Outer Biosciences already stripped of names and contact details by the organizations supplying it. The company pays cost-recovery fees to those suppliers instead of purchasing tissue outright, Polansky said, and it spent roughly two years building the logistics needed to move a sample from the operating room to its lab within a few hours, while it is still living.

A proprietary support system keeps the sample fed with nutrients and clear of metabolic waste once it arrives, which is what stretches its usable life so far past the norm for excised skin: ordinarily a matter of days, long enough for acute toxicity checks but not for slower changes like collagen remodeling or shifts in pigmentation. Outer Biosciences says its system holds tissue viable for close to a month while keeping what Polansky called its “day-zero epidermal, stromal and immune-associated molecular programs” largely intact, so month-old tissue still resembles day-one tissue, if imperfectly. Researchers use that extra time to deliberately damage the tissue with UVB light and observe how its stress signals, immune activity, and repair process evolve across the following weeks, an approach Polansky frames as watching an injury unfold rather than curing one.

Polansky credits the AI layer with a sharp jump in discovery speed. An earlier “brute force” phase, mining scientific literature and testing natural compounds, produced only a couple of leads over roughly a year and a half. The AI-guided version now turns up a new candidate on a rough six-week cycle, and the pipeline currently holds six active leads plus several dozen additional hits, four of which Polansky said are on track to become commercial products.

The persistent bottleneck in biological modeling has rarely been compute. It has been the shortage of data reflecting how a living human system actually behaves, as distinct from a mouse, an immortalized cell line, or a synthetic organoid standing in for one. Outer Biosciences is betting that a smaller, expensively assembled dataset of genuine human response, processed on-premise rather than in the cloud, compounds faster over time than a larger but weaker proxy dataset ever could. It is also why, in Polansky’s framing, there is no “biology internet” to scrape: every data point has to be generated tissue sample by tissue sample.

Because Outer Biosciences develops cosmetic ingredients rather than drugs, it does not need FDA clearance. Its candidates instead get assigned a standardized industry name and go through safety evaluations built around OECD standards, the Paris-based body whose member countries accept each other’s properly conducted studies. The company plans to license finished ingredients to beauty and pharmaceutical partners rather than launch its own product line. It has raised roughly $23 million from investors including Wing Ventures, Initialized, and Polansky’s own firm, Hawktail, and employs 19 people. A Philadelphia rival, Vivodyne, is pursuing a related idea with lab-grown human organ tissue and has raised close to $80 million, a sign investors see promise in real-tissue data broadly.

For teams evaluating AI-driven discovery pipelines, the number worth tracking is that six-week candidate cadence against the company’s own year-and-a-half baseline before AI entered the loop. If that pace holds as the tissue pipeline scales, it argues that proprietary biological data, not model size, is the binding constraint in this category.

TechCrunch’s Connie Loizos reported this profile of Michael Polansky and Outer Biosciences, published August 22, 2026.