The first clue is yeast. Long before machine learning could fill a conference ballroom, Pek Lum was studying a single-celled organism with an almost comic talent for revealing the machinery of life. Yeast is modest company for an ambitious scientist, but that is its charm. Its genetics are tractable, its experiments disciplined, its lessons portable. Lum completed her doctorate at the University of Washington in 1996 after earlier study at Hokkaido University in Japan. The habit she developed there would outlast every fashionable label that followed: look at the whole system, find the pattern, and make the pattern answer to an experiment.
That habit carried her into Rosetta Inpharmatics in 1999, when reading gene expression at scale still felt like peeking through a newly opened door. Biology was becoming data without becoming less biological. The numbers were vast, the context indispensable, and the most useful people could speak both languages. Lum became one of them.
Rosetta was acquired by Merck in 2001. Lum stayed, rose to Scientific Director and helped bring molecular profiling into established drug-development work. Her remit eventually included diabetes, cardiovascular and obesity franchises. In 2004, she was first author on a paper in Cell that used a genome-wide collection of yeast heterozygotes to investigate how therapeutic compounds worked. The details were technical. The posture was enduring: do not ask the data merely to confirm a favorite answer. Ask it to expose the mechanism.
years across genomics and drug discovery
year Auransa was co-founded
decades joining biology and computation
The shape of a difficult problem
After Merck came iPierian, where Lum served as a senior director, and then Ayasdi in 2010. Ayasdi grew from mathematical research around topological data analysis, an approach designed to make the shape of complicated datasets visible. If a spreadsheet is a stack of facts, topology tries to show the landscape those facts make together: neighborhoods, branches, unusual islands, hidden gradients.
Lum joined early and took on two jobs that are often kept politely apart. She was a chief data scientist, responsible for the intellectual integrity of the analysis, and a product leader, responsible for making it useful to someone else. She helped create early business verticals and watched the company grow from a handful of employees to several dozen. In life sciences, she argued that the analytical bottleneck was not simply compute or storage. Researchers had plenty of measurements. They needed better ways to see the whole.
We really want to make a difference to healthcare.Pek Lum, speaking while at Ayasdi in 2014
It is easy to make “data scientist” sound like a person alone with a laptop. Lum's career suggests something more social. An insight has to survive translation. It must move from an algorithm to a biologist, from a biological idea to an assay, from an assay to a development decision, and from that decision to a team with time, money and skepticism at stake. Precision matters at every handoff. So does persuasion.
That dual fluency became founder preparation. By the time Lum and Viwat Visuthikraisee started Auransa in Palo Alto in 2014, she had worked inside a genomics pioneer, a large pharmaceutical company, a regenerative-medicine biotech and an enterprise software startup. She knew what each institution could do. She also knew where information tended to fall between them.
There was another lesson in that tour of institutions: scale changes the question. A research group can pursue a promising signal because it is interesting. A software company must make the signal legible and repeatable. A pharmaceutical organization must decide whether it can become a program. A startup must do all three while explaining why this trail, among hundreds, is worth following now. Lum had seen each version of the argument from the inside.
The resulting company was not a clean break from her past. It was a compression of it. Auransa could combine the curiosity of a lab, the computational reach of a software platform and the development discipline of biopharma. The wager was that those pieces would work better in a continuous loop than as services passed across organizational borders.
Begin with the patients
Auransa's central choice is an ordering choice. Rather than begin with a celebrated target and search for a use, the company begins with molecular and clinical data drawn from people. Its platform analyzes patterns across those patient samples, looking for biologically meaningful subtypes and the mechanisms that distinguish them. A candidate is not the opening guess. It is the result of a chain of questions.
This is what “biology-first” means in practice. Machine learning helps organize complexity; it does not grant the complexity permission to disappear. The output still has to face experimental biology. Lum has called herself a “drug hunter,” which is a more vivid description than chief architect. A hunter reads traces, knows the terrain, discards false trails and waits for evidence before moving.
In 2019, Lum described Auransa's engine as a way to reveal heterogeneity inside a disease. That word, heterogeneity, is where much of the company's strategy lives. A diagnostic name can gather people whose underlying biology differs. Average them too quickly and the useful distinctions blur. Preserve the distinctions, and a subgroup may suggest a mechanism worth pursuing.
The approach produced an internal pipeline and partnerships, then crossed a threshold that matters to any discovery company: an internally generated program moved from computation and preclinical work into clinical evaluation. The company announced regulatory clearance for its lead program in 2022, the first patient dosed in a Phase 1 study in 2023, and the first patient in Asia dosed through a partnership in 2024. Those milestones do not settle the scientific question. They show that the question traveled.
Resilience, without the theater
Founder stories often treat persistence as a form of dramatic weather: storms endured, doors kicked down, destiny announced. The description of Lum offered by UCLA professor Xia Yang is more practical. Yang writes of resilience, positive attitude, flexibility, team spirit and genuine care. When Yang reaches a difficult decision, she asks what Lum would do. The answer is not always to refuse. Lum's example, she says, is knowing when to stand firm and when to compromise.
That is unusually apt for translational science. A founder must hold a scientific thesis tightly enough to protect it and loosely enough to let evidence revise it. Teams need conviction, but experiments are unmoved by charisma. The work demands a personality comfortable with being corrected in public by a result.
Lum's public manner matches the description. She talks about collaborators by name. She frames company milestones as collective work. Her excitement is most visible when discussing the continuity between tool and outcome: a computational engine she helped create, a piece of biology it surfaced, an experiment that made the finding more real. There is pride, but it is pride in the chain holding.
The clever idea gets attention. The translated idea gets used.
Her advisory work extends the same instinct. Lum has advised the Michael J. Fox Foundation for Parkinson's Research, Resolution Bioscience and Sequence Bio. In each case, the useful currency is not abstract enthusiasm for data. It is judgment about where a dataset can change a decision, how a model should be tested, and what evidence deserves the next dollar and month.
A long apprenticeship to the obvious
In February 2026, Lum returned to the academic lectern for the Stivala Lecture at Stevens Institute of Technology. Its title, “Patients to Discovery,” compressed her operating system into three words. Patient-derived molecular information can reveal causal pathways, clarify subtypes and point toward opportunities that simplified models miss. The future she described did not begin with a machine. It began with people.
The title also makes sense as autobiography. Lum's career has moved steadily toward the point where computation, human biology and experimental drug development can share a room. Yeast genetics taught her to respect systems. Rosetta and Merck taught her how molecular evidence enters a pharmaceutical pipeline. Ayasdi taught her to turn mathematics into a product. Auransa gave those lessons a single address.
There is something pleasingly unfashionable about the sequence. It took time. Expertise compounded before it branded itself. Lum was practicing machine learning in biology when the subject still required an explanation, and she kept working after the explanation became a slogan. The advantage was never simply arriving early. It was staying long enough to develop taste.
Taste, here, means knowing which complexity must be preserved, which pattern deserves a laboratory test, and which result is strong enough to move. No model can relieve a founder of those choices. It can make more choices visible. Lum's work is built around that distinction.
So the clue remains yeast: a small organism, a large system, a patient investigator. The tools have changed, the datasets have swollen, and the job title now includes CEO. The question underneath is familiar. What is biology trying to say, and what would count as a fair test? Pek Lum has made a career of listening closely enough to ask it again, better.
Continue exploring