Mendelspod Podcast

Stanford physicist and bioengineer Polly Fordyce has a big vision. She’s attempting to measure protein function at the scale at which we learned to measure DNA. We can sequence proteins. We can increasingly predict their structures. But we still have a surprisingly difficult time measuring what proteins actually do.
Fordyce wants to change that. Her lab is developing ways to measure protein folding, binding, kinetics and function at a massive scale, using the quantitative language of physics. Her ambition is not simply to create more protein data. She wants measurements good enough to make biology more predictive.
Her favorite analogy is weather forecasting. Better computers and better models helped transform our ability to predict the weather. But so did thousands of weather stations around the world making standardized measurements of temperature, wind and precipitation. Biology now has extraordinary computational power and increasingly powerful models. Fordyce thinks it needs the equivalent of those weather stations.
Last year, Schmidt Sciences awarded Fordyce a Polymath Award worth up to $2.5 million to pursue that idea. Her lab has developed a new bead based technology that could allow ordinary laboratories to make high throughput measurements of protein function. Fordyce hopes scientists around the world will contribute those measurements to a new open resource she calls the Functional Protein Observatory.
The implications go well beyond building a database. Fordyce describes recent work from her lab on a protein involved in cancer and developmental disease. After making hundreds of thousands of measurements across human variants, the researchers found that the prevailing model for how drugs act on the protein may be wrong. The drugs appeared to stabilize a previously unseen form of the protein that was only partially closed. That could help explain why drugs designed around the old model have struggled. And it shows what can be discovered when scientists measure how proteins actually behave rather than relying on a static picture of their structure.
AI makes the project more timely. Computational models can now propose proteins and mutations far faster than scientists can experimentally test them. Fordyce believes that gap can be closed. Her new platform can go from receiving a library of DNA to functional measurements within 72 hours. This opens up the possibility of a continuous cycle in which AI can propose, experiments test, and the measurements make the models better.



This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.mendelspod.com/subscribe

Mark Van Oene has just taken over as CEO of Pacific Biosciences, and Mendelspod is his first interview in the new job. He began his career as a scientist studying cystic fibrosis genetics at Toronto’s SickKids, then spent fifteen years at Illumina, rising from its first sales rep in Canada to Chief Commercial Officer. He joined PacBio in 2021 as COO and has spent the past five years helping build the product portfolio he now inherits as CEO.
He joins us today with a clear message: PacBio is choosing long reads. The company has moved on from its Onso short-read platform and is concentrating its resources on making HiFi sequencing cheaper, more scalable and the routine choice in certain clinical applications.
With the new SPRQ-Nx chemistry, PacBio has brought the list price of a HiFi whole genome down to $345. For Van Oene, that changes the problem. “It’s becoming less about economics for me right now.” The bigger constraint is scale, and PacBio expects a new high-throughput system next year.
The opportunities he keeps going to are in the clinic. In rare disease, long reads can consolidate multiple tests while revealing parts of the genome other approaches miss. He sees that eventually leading to long-read whole-genome sequencing at birth: “If you don’t know what you’re looking for, let’s use the most comprehensive analysis you can get.”


This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.mendelspod.com/subscribe

This is a free preview of a paid episode. To hear more, visit www.mendelspod.com

10x Genomics has been remarkably good at figuring out where biology is going next.
In this wide-ranging conversation, co-founder and CEO Serge Saxonov explains why. Despite decades of genomics, “we still understand very little biology.” For 10x, the way forward has been to build better tools for measuring it and to work backward from the biological questions rather than forward from a particular technology.
That philosophy helped drive the single-cell revolution and 10x’s move into spatial biology. Now the company has launched Atera, its new spatial platform designed to measure the whole transcriptome with single-cell sensitivity and at much greater scale. Customer shipments are expected to begin later this year. Saxonov argues that spatial is becoming something much bigger than another research niche. It’s a way of bringing molecules, cells and tissues into the same view.
Going beyond RNA? Saxonov discusses 10x’s recent moves into proteomics and why he believes binder-based approaches currently offer the best path to scale. And he says 10x is investing directly in clinical diagnostics, pointing to therapy selection in oncology and autoimmune disease as two early opportunities. Getting closer to patients, he says, should also help 10x understand what its next generation of technologies needs to do.
What would convince Saxonov that all this is succeeding? Not another instrument or another omic. It would be a measurable improvement in clinical-trial success.
“The next decade is going to be wild,” he says.

This is a free preview of a paid episode. To hear more, visit www.mendelspod.com

Those of us in biology tend to think that chemistry is rather straight forward. But today’s guest says there’s quite an art to synthesizing a molecule.
A few years ago, Lee Cronin, a chemistry professor at the University of Glasgow joined Mendelspod for a sprawling conversation about the origin of life, alien biology, assembly theory, and his conviction that chemistry could become programmable. Now Cronin is back, and that last idea has become a company that now has a “chemifarm.”
First let’s look at the problem. It is easy to assume that once a drug company has designed a promising molecule, the chemists more or less know how to make it. Not so. Chemistry remains surprisingly artisanal. A molecule that looks perfectly plausible on a computer may require too many steps to synthesize, depend on unstable intermediates, or require reactions that simply aren’t known. And chemists may not discover that until they’ve spent considerable time trying.
This becomes an even bigger problem in the age of AI. Algorithms can propose vast numbers of new molecules, but which ones can actually be made? And if one can’t, is there another molecule nearby in chemical space that could perform the same function but be dramatically easier to synthesize?
That’s the problem Chemify is trying to solve.
“You cannot discover what you cannot make,” Cronin says in today’s interview.
He spent 15 years developing χDL, a programming language that reduces the work of chemistry to a set of basic operations that machines can execute. Chemify’s software can then work backward from a desired molecule to determine a possible route for making it. Its robotic systems execute those instructions in the physical world, including chemistry requiring unusual temperatures and conditions. And its Chemifarms bring large numbers of these systems together so the process can be repeated at scale.
The result is something Cronin calls a chemistry “hyperscaler.” A pharmaceutical company might bring Chemify a molecule proposed by its own AI system. Chemify can ask whether that molecule is realistically makeable, develop a route to it, physically attempt the synthesis, and feed what happened back into the system. If the molecule isn’t practical, the platform can suggest alternatives. Every success—and importantly, every failure—adds information about what parts of chemical space are actually accessible.
Cronin’s ambition is enormous. He wants Chemify to become a sort of utility, the “AWS for all drug discovery companies.” Not another company competing to discover the next drug, but infrastructure that allows pharmaceutical and AI companies to turn digital ideas into physical matter.

There are very few genomics projects with the level of ambition of the NIH All of Us Research Program. The latest release includes data from more than 747,000 participants, 535,000 whole genomes, 480,000 electronic health records, and—for the first time—long-read sequencing, proteomics, transcriptomics, and a large collection of information extracted from clinical notes. More than 24,000 researchers are now using the resource. But the program is also arriving at an important transition: roughly 80 percent of its original ten-year funding runs out this year.
So what happens when a massive national research experiment begins to come of age?
Josh Denny, CEO of All of Us, joins us to talk about what the program has accomplished, what researchers are beginning to learn from the data, and what comes next. We discuss the extraordinary scale and diversity of the resource and the growing use of genetics alongside electronic health records and other forms of health data. All of Us is beginning to move from building infrastructure toward producing discoveries that could affect patient care.
Denny explains why the program’s diversity is not simply a matter of representation but a scientific necessity. The project has already identified roughly 1.3 billion genetic variants, including more than 275 million that had not previously been observed. That diversity becomes even more important, Denny argues, as medicine moves toward increasingly personalized predictions and treatments.
“We are capturing such a richer population and so much more kinds of data that we can’t actually reason through it as humans,” he says. “If the data underneath it are highly biased and not representative, then we’re going to make the wrong conclusions.”
We also discuss All of Us as a platform for a much broader picture of human health—from electronic health records and wearables to nutrition, the microbiome, multiomics, environmental exposures, and eventually pediatric data. Denny shares examples of participants whose lives have already been changed by medically actionable genetic results and describes how researchers can build new studies on top of the All of Us population.
What is the next phase of the program and that of precision medicine?
“We’re going to have to redefine our definition of disease,” Denny says. Rather than treating something like type 2 diabetes as a single condition, he imagines increasingly precise descriptions of an individual’s biology, exposures, risk, and response.
After years spent building one of the largest health datasets in the world, All of Us is beginning to show what we might actually do with it.


This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.mendelspod.com/subscribe