Tim Peterson

Evolution is the language model of biology. It’s tractable using cell fitness and genetics.

@LocasaleLab In fairness to them, you probably could have picked a better time to crap on the study like after everyone was done celebrating the win.

Most people dunking are doing ad hominems. They are right to celebrate the KRASi win, just as you are right to say progress could be better. You’re not attacking the person. You’re attacking the system. Can’t understand why it’s so hard for them not to troll as they likely agree with you that the system has serious flaws.

I care deeply about cancer research. I got into this field because cancer affected people around me, including my uncle who passed away from pancreatic cancer. For me, this has always been about combining a drive to understand biology with a desire to reduce human suffering. I want nothing more than for this drug to be successful. I hope the advances that stem from it meaningfully help patients. I hope pancreatic cancer follows a trajectory more like hematologic malignancies such as MM, CLL, CML where outcomes have improved substantially over time, and not what we’ve seen in many solid tumors like lung, colon, or melanoma. I would be very happy to be wrong. This drug also represents very interesting basic science. The originators of this work could very well be deserving of a Nobel Prize in Chemistry. My point is not to dismiss progress or the people working on it. It is to question whether the current trajectory is sufficient to achieve what should be the goal: large scale reductions in mortality. I genuinely hope this is the beginning of something much bigger.

@arcinstitute @MolecularCell @LukeGilbertSF @LukeGilbertSF no LIPN1 amongst top hits? I had a mTOR paper on that.

@arcinstitute @MolecularCell @LukeGilbertSF mTOR!!! awesome stuff Luke!

mTOR inhibitors stop cancer cells from dividing, but they rarely kill them. New work in @MolecularCell from @LukeGilbertSF & lab reveals a mechanism connecting mTOR signaling, cholesterol biology, and ferroptosis that could turn growth arrest into cancer cell death.

@jrkelly @Ginkgo Super impressive congrats!

I don't I'm ever going to get tired of looking at @ginkgo's autonomous lab running a lot of different experiments at midnight. Let us help you kill the bench!

@AlexanderKalian Why did AlphaFold work?

@AlexanderKalian 🎯 Evolution is 10^40. Modern AI is 10^14. Its like a grain of sand vs all the sand on all the beaches on trillions of earths.

Here's one for the "AI will solve biology" crowd... I calculated an estimate for the minimum size of training data required, for AI to have any reasonable shot at "solving" drug discovery. (Not even wider biology - just AI drug discovery alone.) The results are terrifying... ~100 PB of data required. This alone needs a few data centres to store. ~1 trillion molecules in the dataset. Current open-source databases only have 123 million molecules, i.e. 0.012% of what is required. ~3.8 quadrillion bioactivity datapoints. Currently, the largest public database has 295 million bioactivities, i.e. 0.0000076% of what is required. And collecting this data would require... ~21,000 years of continuous research by 1,000 NCATS-style bio labs armed with high-throughput robotics. There are currently only ~100 labs of this calibre in the world. ~$3.8 quadrillion in research costs (not counting the building, set-up, or maintenance costs of the labs). For reference, McKinsey has estimated global pharma R&D expenditure to reach $350 billion by 2029. The research cost for gathering the data for AI to "solve drug discovery" is hence 11,000x higher than this. And even if in the future we build a huge global network of 1,000 giant fully automated bio labs, working 24/7 on this and with much higher efficiency, we can estimate ~10.6 years of non-stop work and ~$38.8 trillion in operational research costs (not counting lab set-up or maintenance). Whichever way you look at it, it currently requires vastly more money than the entire global economy is worth, as well as 21,000+ years of constant research work. I haven't outlined my math here (it is very long and technical, for another post) - but this is all subject to very generous assumptions. The true scale, time, and cost would likely be much much higher. In future, innovation may be able to lower this cost down into the ~$38.8 trillion range (i.e. "only" 1/3rd of the entire projected global economic output of 2026), plus 10+ years of continuous automated lab work. It's not feasible. And this is just for reasonably coming close to "solving" AI drug discovery for small druggable molecules. This is before we discuss other areas of pharmacology - such as nanoparticles, antibody therapies, and gene therapies. This is also before we discuss other topics in wider biology - such as synthetic biology, genomics, epigenetics, neuroscience, a rigorous solution to protein folding, nanorobotics, diagnostics, medical imaging, surgery, virology, parasitology, ecology, evolutionary biology, astrobiology etc.

I’m moved by evidence that if a patient thinks a treatment will work, it is more likely it actually will. Lots of data on that. It’s my understanding currently a clinical trial patient has no idea what’s happening with other participants in the trial. However if there’s a prediction market with money behind it that the patients can see, I worry it will change their belief on whether the drug is working and thus actually affect the underlying biology. So I’m actually most concerned about screwing up whether we learn if the drug works or not.

Watch what @NicholasZaorsky says about EndpointArena.com. A very sharp critique of prediction markets in clinical trials. Well worth the 4 minutes 👇 youtube.com/watch?v=MfuyQa…

Nice piece by @ElliotHershberg: AI versus Eroom's Law. Grateful for h/t @NOETIK_ai & @lifeschemistry. Our thesis is the biggest cost of new medicines is clinical failure. If we can predict clinical response, less failure creates the most value. centuryofbio.com/p/eroom

@ShaneTuttleNCAA I can see college salaries flipping pro salaries because people care more about their college team

UConn G Braylon Mullins will hold the first ever “NIL Auction” to determine what his next team will be with the minimum bid starting at $6M. Mullins intends to return to college and wants to be with the team that is willing to pay him the most.

@endpointarena Cool site love the idea!

Quick run down of the features on Endpoint Arena, the prediction market for clinical trials. the home page has all the trials:

Japanese gym equipment is extremely unconventional. Instead of training just hypertrophy, their machines often strengthen through mobility. They also have 5x the amount of centenarians than the US. I think they’re onto something.

@jeffyfish9 @ResearchHub Clever idea! Great idea Jeff and team!

If you deposit $RSC onto @ResearchHub - you'll start earning yield to deploy towards funding research projects. Re-imagining how endowments operate!

@SVG__Collection Need more of this

Herbie Hancock turning ‘Cantaloupe Island’ into a groove you can’t escape.

@tanaydesaii yes

Scientific breakthroughs often come from “outsiders”, people too naive to respect the consensus. AI will empower a new generation of builders and citizen scientists to be this much needed outsider voice. It’ll completely disrupt the current paradigm of risk-averse, hyper-specialized consensus research designed for citations. Example: Opus 4.6 wrote me a 50-page biology textbook so I could build in a domain I had no formal training in ↓

Neurodegeneration and cancer are highly inversely correlated. We all have either prone to die or prone to survive genomes. I think the stat is cancer is 21x less likely in Alzheimer’s patients and vice versa. The general idea is to restore survival mechanisms in Alzheimer’s prone people. Easier said than done but mainly giving pro-cancer treatments would seem to help. Not joking.

Why focus on Alzheimer's? It's a terrible disease, it is getting worse as populations age, and the complexity of the disease is a good fit for AI. Alzheimer’s affects not only the millions of people diagnosed, but their spouses, children, and other caregivers who support them. The disease places immense emotional and financial strain on families. Modern medicine has made more of a dent elsewhere... 🧵

@realAtlasPress CS Lewis is GOAT. All virtues need courage to be real.

Potentially my favorite C.S. Lewis line of all time

YouTube is such rat poison for my kids. the correlation between time spent and subsequent abysmal behavior could not be more clear. Total ban is difficult but we’re getting close.

I fixed my sleep by stopping tracking it. Turns out this millennial old advice was right.