A brief update to yesterday’s post notes that there are still people who care about the true meaning of epigenetics, and even call themselves theoretical biologists. Note that the Institute for Systems Biology is not some drive-by operation, and indeed is the home of this year’s winner of the Nobel prize in medicine. There may be hope yet. (ᔥJeffrey West, on X)

I have been using OmniFocus since 2016 and from the very beginning have kept a running list of blog post ideas which I almost never use. ⊕ “Write about Taleb’s VC quote” says an entry from October 11, 2024. More than a year later I did write about it, but not because I saw it on the list and have in fact only just now realized that it was on the list in the first place. The oldest active entry is from August 15, 2021: “Write about theoretical biology”. The second-oldest is from four days later: “Write about Waddington’s epigenetics”. This was a few months before I had read any of his books, so maybe it was just mine discovering what Waddington did? In any case, consider this post as a way to cross both of these tasks off the list.

And yet again, the writing is not prompted by any list, but rather by this question on X — what are the major breakthroughs in biology that were idea-driven arguments based on existing data — which duly reminded me of CH Waddington (or, as iOS 26 autocorrect misspelled it just before I had hit return, “CH Washington”). Waddington, a proponent of theoretical biology as a parallel to theoretical physics, organized symposia in the late 1960s on the topic. Alas, it never took off. He died in 1975, age 69, just in time to see research funding for experimental biology skyrocket making everyone an experimental biologist. The theoretical part is now mostly mathematics: see, for example, the Mathematical Oncology newsletter, but what Waddington proposed was not really maths. Interestingly enough the man behind the newsletter, Jeffrey West, has co-authored a paper with Taleb that was very Waddingtonian, with a recent follow-up and a whole book (which I am yet to read).

For an example of what Waddington wrote about see his most well-known work: the epigenetic landscape, proposed before we even knew what genes were. To me these were incredibly useful when thinking about differentiation of complex cells and how it can go sideways. It is also incredibly annoying that the term epigenetic has been hijacked by molecular biologists to mean solely chemical changes to DNA and adjacent proteins which are more likely than not merely a sideshow to what really controls gene expression (3d structure, mRNA, other genes, i.e. everything that goes into a gene regulatory network). Ask a doctor what epigenetics means and the first thing they say will be acetylation and methylation, and if they are oncologists they will talk about “epigenetic drugs” whose job is to inhibit methylation (“hypomethilators”), or what not. I would wager that GLP1 inhibitors like Ozempic are more epigenetic than the most active hypomethilator, but I may as well go after windmills.

Now, the person who asked the question that kickstarted this thinking is the founding editor of Assimov Press which is a charming publication about science and scientific progress. I hope his asking questions will lead to more writing about what happened to theoretical biology and that I’ll learn more about people who carried the flame (or, more likely, rediscovered the concept after everyone forgot about poor old Waddington).

Update: Dr. West has pointed me to the work pf Sui Huang from the Institute for Systems Biology who has tried to bring to terms the two different meanings of epigeneticts with explicit tie in to GRNs. I am sure that very paper is where I got the notion from, but have of course completely forgotten about it. Thank you, Jeff!

The first note is on quickly estimating the 95% confidence interval of an event rate when there are no observed events: if you observe n patients, and none of these patients have the event, then a 95% confidence interval for the probability of the event goes from zero to 3/n (source, with more mathematical detail than I care for). So, if you treat 5 patients and none of them respond, the true response rate could still be as high as 60%. Note that there are many drugs on the market now approved for response rates much lower than 60%, possibly because of the flipside of this calculation (5 of 5 responders could still mean that the true response rate is “only” 40%) combined with some persistence on the part of the developers. But are some drugs dropped too quickly? Probably, which increases the urgency of making clinical trials easier and cheaper to run.

Another implications is that in your standard 3+3 dose escalation design, where you go up in dose if the first 3 study participants don’t experience a dose-limiting toxicity, the 95% confidence interval of the DLT rate at that dose level is still 0 to almost 100%. So, the trials we are running aren’t giving us good enough information. Yay!

The second note, much les philosophical, is that there exists and online tool called reference extractor which can go through a document and extract all Zotero and Mendeley references from it for export into a variety of formats. It can also select those references in your Zotero library, which is life-saving for a slob like me who keeps his references haphazardly strewn across dozens of subfolders. This way anyone who asks can get a neat export, files included.

• David Li: lots of discourse on here recently on how to get FIH clinical data in US cheaper, faster, and more competitively with China. “One thing I have not seen folks talk about, which to me is the most obvious, is - wait for it - the intense competition that breeds insane levels of hard work in China.” From a clinical trial perspective, let’s not forget how receptive the audience is to clinical trials, and how deferential to the doctor/investigator’s opinion. (ᔥRuxandra Teslo)
• John Collison: Dave Ricks has been at @EliLillyandCo for 20% of its 150-year history. He came to the pub, poured his own Guinness, and gave us a 2-hour state of the pharma union. Fear not, there is also a YouTube video of the conversation.
• Michael Eisen: Needed a word, so I coined one. The word is chrysotactic — being drawn to gold, wealth, luxury, etc. — which is derived from Greek but I have to say that the Latin aurotaxis has a better ring to it. You will never guess to whom it relates.
• David Weigel: How Elon Musk’s Changes to X Made Our Discourse Far Stupider. This is not on X but rather an article in Talking Points Memo and I completely agree. And yet, people worth listening to continue using it and as long as that X is the only place where they write I will continue to follow. Perhaps.

• Ruxandra Teslo: What will it take for AI to change drug discovery?. Producing boatloads of hypotheses won’t be enough, and would even set the field back. Good stuff. Again, China is eating America’s lunch here so something better change, and soon, but it is unlikely LLMs will be of immediate help.
• Chris Arnade: Asian style materialism. And if you think, well, maybe it is time for one civilization to sunset so that another one will rise, observations like Arnade’s should make you pause. The culture there seems rather bleak. And I consider him a rather impartial observer.
• Sarah Kliff for the NYT: The ‘Worst Test in Medicine’ is Driving America’s High C-Section Rate. It is about fetal HR monitoring, a particularly salient topic right now. It seems to be the Swan-Ganz catheter of obstetrics, thankfully much less invasive.
• Anish Koka on X: In the wake of the hubbub on novel promising gene therapy approval for Huntington’s disease, this video on the Sarepta debacle is a must watch. Can’t be… I am sure that everything was on the up-and-up.

Where are all the trillion dollar biotechs? asked Lada Nuzhna in her rarely updated blog. I will paste only the conclusion but the entire post is thoughtful and well-documented:

Drug development, like any other industry, is greedy - it addresses the most tractable diseases with the biggest outcomes first. Genetic targets, clear biomarkers, and one-pathway wins gave rise to the biotech boom of the 70 and 90s, when recombinant insulin, monoclonal antibodies, and early gene therapies created a sense of an endless frontier. Unlike with other industries, reinvesting capital from those early wins back into the ecosystem didn’t accelerate industry’s progress – we’ve been on a reverse trend for a while now. Today, remaining problems resist the very playbook this industry was built on.

Most industries have eras when progress stalls before a new paradigm unlocks scale again. Electricity needed transmission grids, computers needed operating systems, and aviation needed jet engines. For biotech, whether the shift will come from new modalities, new regulatory frameworks, or entirely new ways to validate efficacy in humans is not yet clear, but we can, perhaps, outline the boundaries within that future will exist: manufacturing and trials should get cheaper with each run, regulations should become more adaptive, approval frameworks should increase and not decrease in variance, and new therapeutic modalities should focus on unlocking new biology, not just producing slightly better iterations on problems we already know how to solve. Until those new paradigms take hold, building a trillion-dollar biotech will remain caught in Lewis Carroll’s logic: running as fast as we can just to stay in place, and twice as fast to make any real progress.

Note that “trillion-dollar biotech” is (hopefully) just shorthand for a company that produces truly world-changing drugs, and is rewarded accordingly by the all-knowing all-seeing Mr Market to reach a trillion-plus valuation. But if you put dollars first and benefit to humanity second, would that not perhaps contribute to these Alice in Wonderland dynamics? Maybe it’s the gold-digging approach to this decades-long gold rush that caused the shovels to become so expensive, maybe even more valuable than hitting gold. More than that: hitting gold — i.e., developing an effective drug — in this topsy-turvy world can even get you punished.

As Kyla Scanlon stated so succinctly, it is a casino economy now. In biotech it isn’t just now but from its very inception, as I have recently learned, and surely there are downstream effects in this approach to drug development. Again, Nuzhna’s blog post is exceptionally well-written and researched but maybe just maybe the problem deserves to be reframed?

• June 25, 2024: Inside the controversy over FDA’s recent gene therapy approval
• July 18, 2025: Analysts demand transparency after Sarepta’s roundabout disclosure of 3rd patient death
• July 18, 2025: FDA Requests Sarepta Therapeutics Suspend Distribution of Elevidys and Places Clinical Trials on Hold for Multiple Gene Therapy Products Following 3 Deaths
• July 30, 2025: Prasad Resigns From Top FDA Post Amid Fallout Over Sarepta Dispute.
• August 7, 2025: The Sarepta Scandal: Laura Loomer, Vinay Prasad, and the history of pharma’s latest attempt to reassert control at Trump’s FDA
• November 3, 2025: Sarepta’s Duchenne confirmatory trial fails, but biotech will ask FDA for full approval anyway

All this for drugs that cost millions of dollars per dose from a company with $2B in revenue. Neutral people in the know have their opinions too. Know me by my enemies indeed.

The Billion-Dollar Molecule tells the story of the late 1980s and early ’90s world of biotech. The only change since then has been that ⊕ Well, there is one more difference. One billion United States dollars in September 1989 is 2.5 billion of 2025’s USD. one no longer faxes a manuscript and sends supplemental materials by snail mail when submitting to the journal; the personalities, incentives, tradeoffs and challenges are all the same.

Also typical of biotech, and sobering, is that 95% of industry-led research Werth described in the book did not matter: for all the lofty ideals of rational drug design espoused at road shows in investor slide decks, Vertex would chase one fad after another hoping for a hit. Once it got one, a truly life-changing set of drugs for cystic fibrosis, it had burned through so much money and became so profit-driven that it actively blocked low and medium-income countries from developing generic versions. Score for big pharma, which has no qualms about giving away drugs where they are needed.

The story was engrossing enough for me to tolerate Werth’s pulpy writing style, full of adjectives for tortured scientists and smoke-filled rooms. One could easily imagine it serialized on Netflix with distinct chapters, and it had indeed been shopped around, apparently without success. That last link ends with a side note that a movie about Theranos was also planned, and the juxtaposition is apt: more than once Werth notes the dramatic discrepancy between what Vertex management tells investors and the ground truth in the labs. There is certainly a difference between the sociopathy of Elizabeth Holmes and the goings on at your typical fly-by-the-seat-of-your-pants private startup, just not as large as you may think. This is, after all, why most of them go bust.

• Kaitlyn Tiffany for The Atlantic: A ‘Death Train’ Is Haunting South Florida. Dream of the next generation of railway turns into a bloodbath because Americans forgot that train tracks at grade level with other traffic is bad bad bad. So bad that Robert Moses tore down half of New York to fix it. But that was almost a century ago and we are a forgetfull species. (ᔥTyler Cowen)
• Tim Urban: Tales from Toddlerhood. An accurate account, and a particularly salient one for me right now. If you think the second baby gets ignored relative to the firstborn, wait until number three comes along. The main character syndrome of the oldest siblings is real (at least to those of us who are not).
• Lisa Woodley: The Year I Left Design. What happens when you let your career drift, and how to get it back on track. Applicable to more than design. (ᔥGina Trapani)
• James L. Olds: “What Grant Reviewers Actually Look For (and What They Ignore)”. He recommends storyboarding your grant proposals. I would trust his advice, but I also — and please don’t interpret this politically — get the urge to burn that system to the ground because it is a narrative fallacy-producing machine.
• Derek Lowe: mRNA Vaccines and Immuno-oncology: Good News. It seems that LNP-encased mRNA by itself, regardless of what it encodes, may imrove efficacy of immunotherapy for cancer as it “stirs up” the immune systen. Big (and indeed good!) if true. But Lowe ends the post, as he did many of the prior ones, with a tirade against the current HHS, NIH and FDA leadership for spreading doubts and fears about mRNA research, forgetting that (1) much of those doubts and fears were about “overstimulation” of the immune system and the related side effects, such as myositis, which the above (positive!) findings further stoke, and (2) the original mRNA research famously never received NIH funding and was indeed “fringe” science. There is raising of legitimate concerns and then there is performative posturing and Lowe’s writing has shifted firmly to the latter.
• Nick Maggiulli: All the Money, None of the Satisfaction. On people continuing to be stingy even after being firmly in the top 1% of income and wealth. I have recently heard a billionaire-several-times-over unironically start a sentence with “If I had money…” so I can confirm that this is a real phenomenon and some may indeed say that the reason they are billionaires is such a mindset, while the hoi polloi drink their $12 frapuccinos. And if those who can afford such beverages are the hoi polloi, what are the people who can not?

• Raghuveer Parthasarathy: Reading Like It’s 1937. The experiment was to choose only books published in a single year and read these in a short period of time. As expected, there were more duds than classics, and with not that many books coming out in 1937 it amount to very few good books indeed. This kind of experimentation does support the claim that we live in a literary golden age, and I’d say it’s because of a combination of high absolute numbers and us being better at detecting edge cases, not necessarily that the relative percentage of classics per year has increased. If anything, it is probably much lower because of all the AI-generated slop.
• Mike Taylor: If you’d built a “tool” that stupid, why would you advertise the fact? The AI tools described are indeed stupid, but remember that the marginal cost for advertising them is 0, and if you find them stupid you were never the intended audience. In that way they very much resemble phone scams and emails from a Nigerian prince.
• David Shaywitz: The Startling History Behind Merck’s New Cancer Blockbuster. A story from 8 years ago, but it was news to me that pembrolizumab had a very roundabout way of reaching patients. GLP-1 inhibitors had a slightly less circuitous route, but still took too long to get to where they are. Is this truly the best way to develop drugs? (↬Derek Lowe)
• Cal Newport: Is Sora the Beginning of the End for OpenAI? Probably not, and this is not really an edge case, but as someone who has trouble concluding blog posts (which is why you’re reading a list) I just wanted to pause and admire that last sentence.