Pinned Tweet
Starting companies is hard, and it's a boon when you can learn from how others have done it.
There's way more content on tech than biotech, but I thought I'd put together an overview of the good resources that do exist:
For the big picture, start with @Atelfo blog post ✍️: o
And if you want a fun read that tracks 'finding medicines' from prehistory to today, read The Drug Hunters 📚:
amazon.com/Drug-Hunters-I…
Now you know the system that exists. If you're a biologist, next step is to understand that finding a gene/protein target is just the beginning. 'What to do' is essential, but often things fail because you're able to do it in the right place for the right amount and duration (or even know what those are).
Here I'd honestly ask your fave LLM to walk you through 'Medicinal Chemistry 101' 🤖, covering potency, selectivity, the main modalities ADME, PK/PD, formulation, tox studies, CMC.
Make sure you understand the big differences between small molecules, antibodies, protein drugs, siRNAs, and cell/gene therapy.
After that, I'd work through a couple of concrete examples to see things in (messy) practice:
📚 “The Billion-Dollar Molecule” (and the sequel “The Antidote”) on how Vertex was formed and struggled their way to a first success.
📚 For Blood and Money by @nathanvardi. Same: fun & teaches you a lot about the confluence of science, money, and people that go into biotech.
After that, I'd go peruse @LifeSciVC's blog to understand how traditional biotech investors look at these endeavors - the main thing to pay attention to is what risks he keeps calling out.
Note that Atlas and other 'Boston biotech VCs' are a dominant group, but not the only way companies get funded. The original biotechs (all the ones called 'Gen' something), and the 2010s 'techbio' companies have a different thesis inspired by full-time founders as the scarce resource, though most of the risks are the same.
This post is a good place to start:
lifescivc.com/2025/10/twenty…
Finally, Derek Lowe’s “In the Pipeline” blog offers >20y of 1) commentary on biotech next, which you can now assess with the benefit of hindsight, and 2) more details on medicinal chemistry if you found that part exciting. Here's one starting place:
science.org/content/blog-p…
Speaking of, someone should take his best posts and turn them into a book... 😉
I'll expand each of the topics here into more resources throughout the week, stay tuned.
There's way more content on tech than biotech, but I thought I'd put together an overview of the good resources that do exist:
For the big picture, start with @Atelfo blog post ✍️: o
And if you want a fun read that tracks 'finding medicines' from prehistory to today, read The Drug Hunters 📚:
amazon.com/Drug-Hunters-I…
Now you know the system that exists. If you're a biologist, next step is to understand that finding a gene/protein target is just the beginning. 'What to do' is essential, but often things fail because you're able to do it in the right place for the right amount and duration (or even know what those are).
Here I'd honestly ask your fave LLM to walk you through 'Medicinal Chemistry 101' 🤖, covering potency, selectivity, the main modalities ADME, PK/PD, formulation, tox studies, CMC.
Make sure you understand the big differences between small molecules, antibodies, protein drugs, siRNAs, and cell/gene therapy.
After that, I'd work through a couple of concrete examples to see things in (messy) practice:
📚 “The Billion-Dollar Molecule” (and the sequel “The Antidote”) on how Vertex was formed and struggled their way to a first success.
📚 For Blood and Money by @nathanvardi. Same: fun & teaches you a lot about the confluence of science, money, and people that go into biotech.
After that, I'd go peruse @LifeSciVC's blog to understand how traditional biotech investors look at these endeavors - the main thing to pay attention to is what risks he keeps calling out.
Note that Atlas and other 'Boston biotech VCs' are a dominant group, but not the only way companies get funded. The original biotechs (all the ones called 'Gen' something), and the 2010s 'techbio' companies have a different thesis inspired by full-time founders as the scarce resource, though most of the risks are the same.
This post is a good place to start:
lifescivc.com/2025/10/twenty…
Finally, Derek Lowe’s “In the Pipeline” blog offers >20y of 1) commentary on biotech next, which you can now assess with the benefit of hindsight, and 2) more details on medicinal chemistry if you found that part exciting. Here's one starting place:
science.org/content/blog-p…
Speaking of, someone should take his best posts and turn them into a book... 😉
I'll expand each of the topics here into more resources throughout the week, stay tuned.
8
29
200
79,374
For those still doubting, death and dementia are not cool. Let's fix.
In lieu of flowers, memorial contributions may be made to:
Johns Hopkins Alzheimer’s Disease Research Center Fund
alzresearch.org/giving/
Alzheimer’s Foundation of America Caregiver Resources
alzfdn.org/donate/
Photo Credit: Tommy Garcia
0
5
25
872
Even when you get the balance right, being clear on bottlenecks and enforcing high standards can sometimes sound falsely pessimistic among indefinite optimists.
I just want the optimistic outcome to really happen!
I just want the optimistic outcome to really happen!
Being in Boston again has been refreshing, because talking to my friends here who are working chemists and biologists has again made me realize how AI-pilled the Bay Area is. The problem is that some of the skepticism about AI in drug discovery or biology here is at the other extreme. Finding the right balance is the tricky part.
0
1
9
1,407
A data point from @impetusgrants: we have tried human reviewers and LLM reviewers. I think LLM reviewers are already more capable than any individual human reviewer for many grants, because they can spot issues across statistics, biology, epidemiology and other fields all in one proposal.
But humans are still better at tacit knowledge, knowing all of the unpublished experiments, the motivations of researchers, ideas that seem good on paper...
But humans are still better at tacit knowledge, knowing all of the unpublished experiments, the motivations of researchers, ideas that seem good on paper...
Going to disagree here. On average an AI generated review will be better than a human generated one from the average reviewer especially in esoteric or cross disciplinary areas. I do agree that it is better to read the paper first, do ur own review & augment ur review with AI. 1/
6
3
41
4,549
Biology's long timelines are often brushed aside as something that can be brute-forced away with more compute.
The arguments goes: (1) robots can speed up experiments, (2) enough compute can simulate biology, and (3) smarter AI can discover new interventions from existing data.
I think postponing data collection for future robots/whatever is a mistake.
"Can't robots run the experiments faster?" To some extent, yes! Robots can speed up measurements and run experiments in parallel. But observing how organs or diseases change over time, or how interventions change that course, involves incompressible time. Robots can speed up in order to... wait.
"Can't we simulate, what about AlphaFold?" It benefited from 50 years (!) of protein-structure data. You can't check your simulations without data.
Learning how aging and disease unfold, and what changes their course, requires physiological datasets that can take years to collect, however quickly models improve.
The argument isn't to collect data as usual, ignoring or pausing AI. Collecting data alongside training better models puts us on the most optimistic trajectory where AI is maximally useful for aging and disease.
We (certainly anyone promoting AI cures) should allocate a fraction of the budget to high-quality data collection.
The arguments goes: (1) robots can speed up experiments, (2) enough compute can simulate biology, and (3) smarter AI can discover new interventions from existing data.
I think postponing data collection for future robots/whatever is a mistake.
"Can't robots run the experiments faster?" To some extent, yes! Robots can speed up measurements and run experiments in parallel. But observing how organs or diseases change over time, or how interventions change that course, involves incompressible time. Robots can speed up in order to... wait.
"Can't we simulate, what about AlphaFold?" It benefited from 50 years (!) of protein-structure data. You can't check your simulations without data.
Learning how aging and disease unfold, and what changes their course, requires physiological datasets that can take years to collect, however quickly models improve.
The argument isn't to collect data as usual, ignoring or pausing AI. Collecting data alongside training better models puts us on the most optimistic trajectory where AI is maximally useful for aging and disease.
We (certainly anyone promoting AI cures) should allocate a fraction of the budget to high-quality data collection.
Despite staggering advances in basic science, drug development research is only getting slower. The main culprit? Increasing clinical trial timelines. The median clinical development pathway has increased from 8.4 years to 9.3 years and pipeline creation has slowed 32% even as
This is the point I keep making about the main bottleneck to drug discovery and why AI on its own won't be enough. These timelines aren't going anywhere (ofc AI can be used creatively to ease such bottlenecks -- eg via better endpoints that allow smaller cohorts for trials, followed up for shorter periods of time).
4
7
35
3,555
Martin Borch Jensen retweeted
Mosaic in vivo screens with single-cell readouts map each target's effects across organs before a drug program locks in. @MartinBJensen at #ARDD2026. Photos: agingpharma.org/photos2026
0
4
6
1,430
Martin Borch Jensen retweeted
The two most beautiful words on Wikipedia. Smallpox was the killer of infants. It was the crippler of those who survived. It was endemic. It was inevitable. Smallpox was, and is no more. Humanity stands triumphant over its carcass.
I’ve asked this before but is there any historical event that is like the Opposite of 9/11. Like something so good that happened that it changed the world forever
24
1,352
15,378
141,301
Martin Borch Jensen retweeted
"Alphafold solved this subset of static structure prediction with a dataset that took 50 years to collect."
If we hope AI will help with aging, we should start generating the data it will need now. If we wait until we need them, we’ll be disappointed.
It’s not a question of AI development and instead just collect data. We can take a micro fraction of the overall budget and use it to generate the right data in parallel with everything else.
The next Impetus Grants @ImpetusGrants round we're setting up is a direct path to that. Our goal is $15M to get these datasets started, and we’ve raised ~$9M so far. norn.group/impetus-grants
We already anticipate compute demand and build data centers years before we need them. We should do the same for training and verification data on aging and disease.
If we hope AI will help with aging, we should start generating the data it will need now. If we wait until we need them, we’ll be disappointed.
It’s not a question of AI development and instead just collect data. We can take a micro fraction of the overall budget and use it to generate the right data in parallel with everything else.
The next Impetus Grants @ImpetusGrants round we're setting up is a direct path to that. Our goal is $15M to get these datasets started, and we’ve raised ~$9M so far. norn.group/impetus-grants
We already anticipate compute demand and build data centers years before we need them. We should do the same for training and verification data on aging and disease.
I’m seeing a lot of uninformed takes, like this one, about cell simulation feasibility. I wanted to contribute some facts.
Alphafold works for predicting a subset of static structures (not how a protein structure changes shape under forces) on 1) well-ordered regions (not intrinsically disordered regions or long flexible loops) and 2) on proteins with plenty of evolutionary relatives (no orphan proteins).
Alphafold solved this subset of static structure prediction with a dataset that took 50 years (PDB; 1971 to ~2020) to collect. It did not solve protein dynamics or how proteins fold into their shapes. Those remain unsolved.
Cell simulation is vastly more complex and data hungry than any of this. Exact cell simulation means solving quantum equations for every atom and is totally intractable. But all approximations rely on - and are bottlenecked by - data, not compute.
So the extrapolation from “alphafold worked” to “therefore cell simulation is feasible” assumes far too much.
1
6
25
5,187
Martin Borch Jensen retweeted
The @impetusgrants from @NornGroup are raising to do a round for critical longevity datasets. We need these to enable AI to help cure diseases. Help fund it!
"Alphafold solved this subset of static structure prediction with a dataset that took 50 years to collect."
If we hope AI will help with aging, we should start generating the data it will need now. If we wait until we need them, we’ll be disappointed.
It’s not a question of AI development and instead just collect data. We can take a micro fraction of the overall budget and use it to generate the right data in parallel with everything else.
The next Impetus Grants @ImpetusGrants round we're setting up is a direct path to that. Our goal is $15M to get these datasets started, and we’ve raised ~$9M so far. norn.group/impetus-grants
We already anticipate compute demand and build data centers years before we need them. We should do the same for training and verification data on aging and disease.
6
7
33
3,061
Re. compute vs bio data collection, path is obvious when you consider EV of marginal spend on either:
If a frontier lab put $50M into collecting data that may be gating, e.g. rich multiomics from real human longitudinal trials, that's ~1 day of compute budget.
Such data takes m study initiation.
If pure simulation doesn't need this data, cures delayed 1 day.
If the data guides models to simulate sooner, cures accelerated by months or years.
If the data is strictly necessary, cures accelerated years.
Obvious choice if eg the first AI benefit example in your S-1 is ""profoundly accelerate the rate of progress in biology and medicine, potentially curing diseases that have plagued humanity for millennia ..."
If a frontier lab put $50M into collecting data that may be gating, e.g. rich multiomics from real human longitudinal trials, that's ~1 day of compute budget.
Such data takes m study initiation.
If pure simulation doesn't need this data, cures delayed 1 day.
If the data guides models to simulate sooner, cures accelerated by months or years.
If the data is strictly necessary, cures accelerated years.
Obvious choice if eg the first AI benefit example in your S-1 is ""profoundly accelerate the rate of progress in biology and medicine, potentially curing diseases that have plagued humanity for millennia ..."
2
5
19
1,402
More details here: norngroup.substack.com/p/ask-not-what…
Ready-to-go 501c3 for making the data happen: @impetusgrants
Please get in touch.
Ready-to-go 501c3 for making the data happen: @impetusgrants
Please get in touch.
0
1
5
186
Martin Borch Jensen retweeted
Endless back and forth about whether bottom-up simulation of biology is technically feasible. In almost every other physical domain we are focused on putting sensors everywhere to collect time series data and then using that to model systems. There is a continuous flow of electricity running from your head to your fingertips and toes instructing downstream biological processes and receiving signals back in a loop and we are hardly recording it. Instead, we are obsessing over building virtual cells in abstracted environments that repeatedly achieve little predictive value.
Where do you think the bigger marginal return to time and money is likely to be?
Where do you think the bigger marginal return to time and money is likely to be?
12
8
134
10,725
Martin Borch Jensen retweeted
@Noahpinion @ramez right. But it sounds a lot like indefinite optimism to me. In the meantime, shouldn't we start thoughtfully generating the data distributions that entail organic time and figure out how to build the instrumentation we need to generate more useful data distributions? The instrumentation is likely to require breakthroughs in material science, etc. This thinkism around....muh...just moar compute...line go up...,is sort of a pointless exercise. Let's think deeply about the potential bottlenecks and launch parallel efforts in case compute doesn't in fact solve everything.
I think you'd enjoy: norngroup.substack.com/p/ask-not-what…
amplesa.substack.com/p/ai-for-bio-d…
I think you'd enjoy: norngroup.substack.com/p/ask-not-what…
amplesa.substack.com/p/ai-for-bio-d…
3
1
3
1,232
Martin Borch Jensen retweeted
Agree. Some questions (like how a disease develops, or whether a drug has long term health benefits for patients) can only be answered by following people over years. If we don't start those studies now, then future AI will not have access to that data.
Lots of things we could do: we can follow patients’ health over time, see how they respond to treatment, and even bank blood samples for tests that may become available later. Laboratory experiments that take days or weeks can run alongside those studies.
We’re raising $15M for the next round of @ImpetusGrants to fund work like this, with ~$9M raised so far. More here: norn.group/impetus-grants
Lots of things we could do: we can follow patients’ health over time, see how they respond to treatment, and even bank blood samples for tests that may become available later. Laboratory experiments that take days or weeks can run alongside those studies.
We’re raising $15M for the next round of @ImpetusGrants to fund work like this, with ~$9M raised so far. More here: norn.group/impetus-grants
There’s a really pronounced divide between people who are on the AI side of things and people who are on the biology side of things. Even among people working in AI, the divide persists based on where they’re focused.
The AI side, adherents of scaling laws and the bitter lesson, believe that AI getting better means AI being more capable of novel discoveries in medicine (that go beyond discovering new drugs in well understood targets or speeding up dry lab work etc).
The biology people think the idea of a bitter lesson in biology is kind of ridiculous and that regardless of AI’s capabilities it’s going to be at least 5-10 years until you could hope to reliably simulate human toxicology in silico (right now AI/ML can’t even reliably simulate in vivo studies like binding affinities or functional assays).
Even the most AI pilled of them, who will list all the areas that AI can reduce the time spent on various areas, don’t think the concept of AI discovering novel targets or generating new understanding of causal biology is realistic in the near term. Most, in fact, will readily characterize the assumption that even an “ASI” level model could find cures hidden in our current understanding (without expanding that understanding with significantly more biological data) as patently false.
This was a big part of why we wanted to experience it ourselves by using AI for a (very) small portion of what it can do in biology. It was both mind blowing and revealed some poignant insights into how different the problem of making AI better at something like coding is to the problems of biology.
Even if it unlocked the ability to generate new molecules at a stunning pace, its simulation (in silico) of binding only had 49% accuracy. And if we wanted to improve that, we’d need to feed it back the data from the lab - data that took about 30x as long to get as it took to generate the molecules. And these were tests in a test tube (turn of phrase), imagine how long it would take if we wanted to improve the model with data that required recruiting humans and monitoring their reactions.
Human biology is very complex, and adding a drug - especially a novel drug and doubly so for a novel drug acting on a novel mechanism - means mapping interactions that have nth-order-effects that we don’t have the data to predict. Virtual cells have made advancements enabled by AI, but the perturbations equate to a small fraction of the possible.
This is before you even get into curing disease, especially uniquely human diseases. Brain function involves cellular identities that are unique to humans, meaning animal studies are of little use.
How long does it take to understand what the side effects of chronic administration of a drug in humans are? AI can speed up analyzing the data, or designing the trial or even getting it approved…but it can’t make a 6 month trial complete in 3 months.
I think its important to point out that significant advancements can be made in AI that contribute to biology, while also being realistic about how much it speeds up something as complex as curing a specific cancer or diseases of the brain.
0
3
8
652
Martin Borch Jensen retweeted
I agree.
Robots are the non-biologist's abstraction of how scaling in biology scales. Imagine people in the 1940s pouring money 100x money into automating the vacuum tube factory than inventing the transistor. Imagine putting everything into giant CPU factories instead of inventing the GPU.
We are disproportionally thinking/resourcing innovation in biology as if it's car manufacturing. There is ample opportunity to improve, redefine and scale measurement in front of us.
I recently talked to someone at a robot-first science factory with near billion in funding boasting how they measure ~1000 mutants a week. A team of 4 at Dyno scaled the equivalent assay multiplexing by 25x. 1 person can run it. The robot is scaling the wrong layer.
I'm not against robots and I think they will have the role to play. But I think their impact on biology is overblown, especially on how it scales in performance per dollar deployed. The danger is we become so obsessed with vacuum tube factories that we don't fund the transistors of biology.
Complex physical value chains are somewhat path dependent, once you build a highway instead of a subway, you have to deal with a lower efficiency, higher pollution world. Most current robotic approaches are a crude, and poorly scalable, way to scale biology.
Robots are the non-biologist's abstraction of how scaling in biology scales. Imagine people in the 1940s pouring money 100x money into automating the vacuum tube factory than inventing the transistor. Imagine putting everything into giant CPU factories instead of inventing the GPU.
We are disproportionally thinking/resourcing innovation in biology as if it's car manufacturing. There is ample opportunity to improve, redefine and scale measurement in front of us.
I recently talked to someone at a robot-first science factory with near billion in funding boasting how they measure ~1000 mutants a week. A team of 4 at Dyno scaled the equivalent assay multiplexing by 25x. 1 person can run it. The robot is scaling the wrong layer.
I'm not against robots and I think they will have the role to play. But I think their impact on biology is overblown, especially on how it scales in performance per dollar deployed. The danger is we become so obsessed with vacuum tube factories that we don't fund the transistors of biology.
Complex physical value chains are somewhat path dependent, once you build a highway instead of a subway, you have to deal with a lower efficiency, higher pollution world. Most current robotic approaches are a crude, and poorly scalable, way to scale biology.
Maybe controversial: the conversation about scientific bottlenecks is overfocused on automation. New measurement technologies deserve more attention. You could automate an assay that needs an entire facility – or you could REINVENT the assay so that it fits into just a few tubes.
28
54
351
30,361
Honorable
SF’s Bottom of the Hill will live on under a new owner in 2027, with a longtime patron taking over the business and pledging to preserve its independent spirit, while the original owners will keep the building to help ensure it remains a music venue.
buff.ly/tMHwiZl
0
0
4
676
Martin Borch Jensen retweeted
hope we get a lot more people giving back to sf nightlife / cultural spots in the coming years
SF’s Bottom of the Hill will live on under a new owner in 2027, with a longtime patron taking over the business and pledging to preserve its independent spirit, while the original owners will keep the building to help ensure it remains a music venue.
buff.ly/tMHwiZl
1
2
50
3,400
Martin Borch Jensen retweeted
Today in @NatureBiotech we introduce the "recordanelle", a synthetic organelle that enables transcriptome-wide single-cell RNA recording in human cells. Big thanks to team: @ReinkemeierCD, Marcos Manero-Carranza, et al. 🧵👇
@ETH_en @ETH_BSSE @BRC_CH
rdcu.be/5Pus7dyu63ta
@ETH_en @ETH_BSSE @BRC_CH
rdcu.be/5Pus7dyu63ta
Ever wondered what a cell has experienced in the past? This is in general hard to measure. Here we introduce the recordanelle, to equip cells with a tool that lets them record their own history.
nature.com/articles/s4158…
@NatureBiotech @ETH @randall_platt @ETH_BSSE
3
41
219
20,515