In February 2001, while headlines declared that the human genome had been read, Brian Gilman sent a thank-you note back to Union College. He had graduated three years earlier with a biochemistry degree and was working at the Whitehead/MIT Center for Genome Research, one of the lead groups in the public sequencing consortium. The celebrated Nature paper carried hundreds of names and a civilization-sized claim. Gilman’s message was smaller. He remembered the professors who put a green freshman in a laboratory, let him puzzle through problems and challenged settled thinking. Publication week, he wrote, had delivered “one of the happiest moments of my life.”
It is an unusually revealing origin story for a technology executive. The big achievement was collective; the memory was about learning how to figure things out. Twenty-five years later, Gilman is chief executive of Plex Research, a Cambridge company building AI software for drug discovery. Between those two points lies a career of networks, cloud platforms, laboratory notebooks, enterprise sales and three founder exits. The fashionable nouns keep changing. His problem has barely moved: science produces far more information than a person can comfortably navigate, so how do you bring the useful piece to the surface without losing the trail that makes it credible?
A biochemist walks into the software business
Gilman’s formal training was in biochemistry, but his working life quickly blurred the distinction between a scientist and an engineer. At Whitehead he operated around genome-scale data when “scale” meant something both more physical and more intimidating than it does in a software pitch. The public Human Genome Project had to coordinate laboratories, sequence fragments, computational tools and annotations across institutions. Its 2001 draft offered the first panoramic look at the genome, but it also dramatized the problem that would follow: reading the letters was only the beginning. Organizing what they meant would occupy a generation.
Gilman moved toward tools for the people doing that organizing. Panther Informatics and then SciLink explored scientific information and professional connection. SciLink was profiled in 2007 as a network built to find relationships among researchers. Wingu, co-founded with Nick Encina, approached the mess from inside the laboratory. Its cloud-based system captured documents, experimental records and collaboration in one place. Gilman’s compact description was “Google for research.” The analogy now sounds obvious. In 2011, asking pharmaceutical teams to put research workflows in the cloud was a rather more interesting dinner conversation.
Wingu also exposed an enduring truth about scientific software: the difficult part is rarely storing another file. It is preserving enough context for someone else - or your future self - to understand why the file matters. Its Elements platform became a modular electronic laboratory notebook and was later presented at an American Chemical Society meeting as a teaching tool for organic chemistry. Students could receive procedures, submit work and combine spectra, images, video, spreadsheets and chemical structures. The product was not merely digitizing paper. It was trying to keep the conversation attached to the experiment.
“Trust is the bottleneck.”Brian Gilman, on AI in life sciences
The founder learns the enterprise
A career spent selling tools to scientists produces a special kind of bilingualism. Researchers care about whether a method respects the work. Enterprises care about security, integration, adoption and return. Products fail politely in the space between. Gilman’s later roles took him directly through that gap: cloud products and sales at PerkinElmer, business development at computational and agricultural technology companies, commercial leadership at BioBright and then Dotmatics, and research technology inside Vertex Pharmaceuticals from 2022 to 2024.
The route matters because scientific AI is not being introduced to a blank page. Laboratories already have instruments, databases, notebooks, conventions and expensive habits. A clever model can be accurate and still be irrelevant if it interrupts the way evidence moves through an organization. Gilman has occupied the startup side, the vendor side and the customer side of that negotiation. Public recommendations from former colleagues emphasize the same combination: enthusiasm about new technology, enough scientific understanding to apply it, and a willingness to keep coding until an idea became concrete.
There is also a revealing childhood footnote. Gilman says he began a business at 13 and earned enough to help pay for college. It is tempting to place a miniature chief executive in the scene, but the more useful detail is the early loop: see an opening, make something, discover whether anyone values it. His career has run that loop at increasing scale. Scientific users are famously good at detecting whether a product builder has observed the work or merely admired the market.
The answer is less interesting than its receipts
Plex Research was founded in 2017 by computational biologist Douglas Selinger. The company’s core idea is a “focal graph,” a knowledge graph shaped around a particular research question. Its platform searches across chemical biology, gene expression, genomics, proteomics, metabolomics and other biomedical data, looking for converging relationships. Large language models help with the interface and orchestration. The underlying pitch, however, is almost stubbornly old-fashioned: every result should lead back to experimental data a scientist can examine.
When Gilman announced his move into the CEO role in January 2026, he focused on that traceability. Give Plex a compound structure or gene signature, he explained, and the system can surface support scattered across tens of thousands of biomedical datasets. Scientists can follow the path and evaluate it themselves. He also listed the possible destinations: target identification, safety assessment, biomarker development, precision medicine and disease-indication discovery. Each could support a separate business. Plex intends to put them on one platform.
From fluent AI to decision-grade AI
This is where Gilman’s biography becomes more than chronology. The Human Genome Project was partly a triumph of making a common resource available. SciLink tried to map who knew what. Wingu tried to make the research record accessible in the cloud. Plex is trying to connect the answer to the experimental chain beneath it. Different architectures, same instinct: a result becomes more valuable when another person can find it, inspect it and use it.
Useful is not yet decision-grade
Gilman has lately sharpened the distinction. “AI in life sciences” is no longer, in his view, an interesting conversation by itself. The better question is where AI is trusted enough to change a real decision: which target moves forward, which biomarker earns attention, which study is designed differently, which weak idea is retired sooner. Speeding a workflow and impressing a demo audience are lower bars. A system becomes consequential when a scientist is prepared to alter the next expensive move.
That standard is useful because it resists both evangelism and dismissal. Plex does not ask scientists to abandon judgment. It gives judgment a larger field of evidence to work with. Nor does transparency guarantee that a conclusion is right. It makes disagreement productive. A researcher can challenge the datasets, the relationships or the interpretation instead of arguing with an inscrutable score. In a domain where the cost of a confident mistake is measured in programs and years, inspectability is not decorative ethics. It is product design.
The interface gets attention. The provenance earns permission.The operating lesson behind Gilman’s career
Plex remains a small company. Its public LinkedIn profile lists fewer than a dozen employees. That gives Gilman a familiar founder’s problem in his first act as CEO: turn a broad technical capability into a sequence of specific reasons to buy. The company has begun offering individual licenses alongside work with larger biopharma customers. It is also publishing technical work, demonstrating compound-target identification and taking its evidence-first language to industry events. The ambition is expansive; the sales conversation must be wonderfully concrete.
The through-line offers a practical lesson for anyone building in AI. Do not begin with the magic trick. Begin with the decision and work backward. What would a careful person need to see before acting? Which context would let them reject the machine’s suggestion intelligently? How does the evidence survive the trip from database to interface? Fluency may open the door, but a chain of custody keeps the product in the room.
Gilman’s career began near one of modern biology’s great acts of reading. It has continued through a quarter-century of attempts to make the resulting flood legible. Now the machines can summarize, propose and converse. The next task is less theatrical and more demanding: helping a scientist know when an answer deserves to matter. For a builder who has spent decades making science easier to search, that sounds less like a reinvention than the latest version of the same unfinished job.