Lucas Nivon has spent much of his career staring at folding problems. RNA first, then proteins, then the stranger contortions of a scientific idea trying to become a company. Each can begin with an enormous field of possibilities. Each must somehow arrive at a useful shape.
His answer has rarely been to make the science sound simpler than it is. He has tried to make the machinery around it more usable. That distinction explains the arc from Harvard biophysics to David Baker's laboratory at the University of Washington, from a browser interface for Rosetta to Cyrus Biotechnology, and from proprietary discovery work to the shared infrastructure of OpenFold.
There is an appealingly practical quality to the journey. Cyrus did not begin with a grand declaration that software would replace experiments. It began because a capable suite of protein-modeling tools was difficult for many working scientists to operate. Rosetta could predict structures, redesign interfaces, and help stabilize proteins, but it came with a learning curve steep enough to require years of experience and serious computing. Nivon saw commercial possibility in the inconvenience.
Chapter oneThe scientist learns to notice friction
Nivon grew up in New York City and arrived at Harvard as a biochemical sciences student with enough physics and chemistry in the mix to be named a Goldwater Scholar in 2000. His undergraduate thesis examined the physics of DNA moving through biological nanopores. In 2001, he was selected for a Hertz Fellowship, a demanding graduate award in applied science.
The news produced one of the better clues to his temperament. Nivon told the Harvard Crimson that the acceptance envelope was of such middling size that it could not immediately be sorted into either the rejection or acceptance pile. He added that the fellowship would cover extravagant items including rent, food, and running water. Dry humor is useful equipment for a career in which elegant theories must eventually meet budgets, timelines, and finicky experiments.
“The envelope they sent was sort of middling in size.”Lucas Nivon, on receiving his Hertz Fellowship result in 2001
He stayed at Harvard for a Ph.D. in biophysics, working with Xiaowei Zhuang and Eugene Shakhnovich on how structured RNA molecules fold. The work crossed computational methods with single-molecule experiments. He built an RNA-dynamics modeling toolkit from scratch and published on the thermodynamics and kinetics of the hairpin ribozyme. The pairing mattered. Code could explore a molecular landscape, but experiments still got a vote.
At the University of Washington, in Baker's lab and the Institute for Protein Design, Nivon moved deeper into computational protein engineering. He worked on small-molecule binding proteins, helped improve the stability and usability of Rosetta, contributed algorithm and database projects, and became an inventor on protein-labeling work that later received a U.S. patent. The assignment was scientific, but usability kept appearing inside it.
Chapter twoA browser window becomes a company
The Cyrus project was conceived in April 2014 by Nivon, Yifan Song, and Javier Castellanos, all working in Baker's orbit. The company took its name from Cyrus Levinthal, whose famous paradox asks how proteins fold so quickly when an unfolded chain has an astronomical number of possible conformations. It is a fine name for a startup too: the path looks impossibly large until constraints begin doing their work.
During a 12-month incubation, the founders completed the National Science Foundation's I-Corps program, built a prototype, met investors, and worked through licensing with the university and Rosetta Commons. Their first proposition was direct: put a friendlier graphical interface around complex Rosetta procedures, automate the awkward parts, and deploy the compute in the cloud. On May 1, 2015, Nivon and the founding team left the university for a WeWork in Seattle's South Lake Union.
At launch, Nivon emphasized customer interviews. Cyrus would listen to scientists, identify the Rosetta functions that mattered in real workflows, and support the difficult tasks around them. This was not a scientist reluctantly tolerating product work. It was a scientist recognizing that the interface, training, computing, and support were part of the invention's practical life.
The software found an audience. Cyrus says Nivon brought the company to sales with more than 120 biopharma companies and collaboration on over 30 discovery-stage programs. The exact counts are less interesting than what they gave the team: repeated exposure to the moments when a promising protein project became stuck.
In a 2026 essay, Nivon wrote that outside teams tended to call Cyrus after wrestling with a problem for six to twelve months. Across dozens of assignments, the company learned where computation was helpful, where established methods already worked, and where experimental data was still missing. Customer service became a field study. The consultancy-shaped edge of the business was gathering a map of unsolved work.
Chapter threeThe wet lab gets a seat at the table
Cyrus eventually changed shape. The company moved beyond making software available to customers and toward an integrated system combining computational design with large-scale laboratory screening. In December 2021, it announced an $18 million Series B and acquired Orthogonal Biologics, a University of Illinois spinout built around deep mutational scanning. The acquisition added experimental breadth to a company born in code.
This looks like a pivot only from a distance. Nivon's graduate work had already paired simulation with single-molecule observation. His later argument about protein AI is similarly unsentimental: an algorithm earns its keep by helping solve a specific development problem, not by supplying a fashionable label. In his telling, Cyrus's approach is neither computation alone nor laboratory work alone. It is the loop between them.
That loop changes the kind of company Cyrus can be. Software proposes a tractable set of designs. Parallel experiments test them. The results return as better evidence for another computational round. Instead of asking the model to perform magic, the system asks it to make the next experiment more intelligent, then allows the experiment to correct the model.
- Harvard and Hertz
Graduates in biochemical sciences and begins the fellowship path toward biophysics. - RNA folding
Builds computational tools and works with single-molecule methods during his Ph.D. - Cyrus takes shape
A translational project becomes the Institute for Protein Design's first company spinout. - Computation meets screening
Cyrus announces its Series B and acquires deep-mutational-scanning capabilities. - OpenFold
Nivon helps organize an open-source consortium for molecular AI infrastructure.
Chapter fourBuild the commons, too
OpenFold extends the same accessibility instinct beyond Cyrus. Nivon co-founded the consortium in 2022 with partners from industry and academia and now serves on its executive committee. The group develops open-source systems for molecular structure prediction, including training and inference infrastructure intended to be inspected, adapted, and retrained.
The word “open” can be decorative. OpenFold makes it operational through code, model weights, datasets, governance, and a community of companies and researchers contributing money or technical work. Nivon described the launch of OpenFold3 as an expression of the consortium's “sharing ethos.” The phrase fits his longer record. Cyrus made an academic toolkit easier to reach. OpenFold tries to make foundational AI infrastructure something a wider scientific community can examine and extend.
There is also a useful tension here. Nivon runs a venture-backed company while helping steward shared infrastructure. He does not treat commercial work and an open technical commons as enemies. Companies can compete on the specific systems, data, execution, and products they build while cooperating on expensive foundations that benefit from scrutiny and common standards.
His side interests make the pattern feel less accidental. Nivon's public profile mentions bikes and urbanism alongside protein AI. In 2012, before Cyrus launched, he noticed a gap in Seattle's long-term bicycle-rental market and began Seattle Monthly Bike Rental from his garage with a handful of bikes. It became Pedal Anywhere. He recruited Zach Shaner to lead it and stayed on as an adviser while returning his attention to biotechnology.
Bicycle logistics and protein engineering do not share a laboratory bench. They do share a founder's reflex: notice an awkward system, isolate the missing layer, and make access less painful. Nivon's most consistent subject may not be proteins at all. It may be infrastructure, especially the kind that turns specialist capability into a usable service.
Powerful tools do not spread on technical merit alone. Someone has to build the doorway, the workflow, and the reason to walk through.
The long experimentOne question, asked at larger scale
Nivon's career is coherent without being tidy. Academic simulation led to protein design. Protein design led to cloud software. Software and customer work exposed the need for experimental scale. Company building led back toward a shared, open-source foundation. At each turn, the object grew larger, but the question remained recognizable: how can a difficult scientific capability become usable in the world?
That question avoids two common fantasies. One is that a brilliant algorithm naturally becomes a product. The other is that a polished interface can remove the stubborn physical work beneath it. Cyrus's history argues for translation in both directions. Researchers need better access to computation. Computation needs disciplined experimental feedback. Both need operators willing to care about licensing, interfaces, customer interviews, capital, and the ordinary mechanics of a team.
Levinthal's paradox supplies the company name, but it also offers a final image. A protein does not search every possible configuration before finding its form. Constraints guide the path. Nivon's constraints have been practical ones: what scientists can use, what customers repeatedly cannot solve, what experiments can verify, and what infrastructure a community can maintain together.
The result is not a straight line from researcher to executive. It is a folding process. The scientist remains visible in the founder. The product manager appears inside the algorithm builder. The commercial operator keeps a place for the commons. And the fellow who once joked about funding rent and running water is still occupied with necessities, only now they are the plumbing of computational biology.