Cheyenne

@parmita Cell therapy will be one of the most effective anti aging treatments IMO

“parm if you make a drug, just go into peptides. it’s clearly the next big thing” Cell therapy, not peptides, will be the biggest realized CAGR (2-3x more) in 10yrs. You may not have heard of it. That’s mainly bcs reality takes some time to catch up to science.

I interviewed @geochurch and learnt we're already gene editing humans. You just haven't noticed. 0:00 - Gene Editing Mammals → Humans 8:36 - Germline vs Somatic 14:56 - Modified Humans Are Already Here 18:50 - Enhancing Healthy Humans 25:00 - Aging Therapies vs Cognitive Enhancement 30:20 - Embryo Selection 38:10 - Is US Losing To UAE? 42:33 - Biotech Failures 49:31 - Next Dire Wolf Moment 54:21 - AI x Science 1:02:07 - Synthetizing Entire Genomes

@m_goes_distance Awesome episode. Watched it yesterday

>be me >interview the Harvard professor >ask about gene editing humans >"we're already doing it" >what about germline editing? >"it's the cleaner solution" >ask about modified humans walking among us >"they're already here" >how far away sir? >"tomorrow" >ask why we haven't >[redacted] >shift gears >"US biotech losing to UAE?" >he nods >ask about the next dire wolf moment for humans >"our ancestors wouldn't [redacted]" >thank George, banger episode >I post the episode on X >algorithm nukes it >cry >WTF? link below

@somewheresy Don’t leave once you succeed. Keep at it, keep winning.

Do not, under any circumstances, move back to your hometown. Go to the largest city with the largest market for the thing you want to do and only leave once you’ve succeeded. You can make your 20s harder to have an easier 30s and 40s

We should give the average AI user the ability to run their own experiments, submit their own regulatory filings, etc, etc IMO this is an appeal to heaven scenario. People should be given agency over their own health and treatments, choosing how much risk they are willing to take on.

The professional biotech world can choose to shun people who use AI to get involved in biotech Call them irresponsible. Stupid. Grifters. But know that the result of that choice will be the professionals are marginalized and the AI work goes on without their input

@josiezayner I’m better than Claude because I spool up a legion of Claudes.

Interviewed a bunch of people for a computational biologist role and asked them all why they are better than Claude and so far no one has given a reasonable answer

@sokrypton @mkoeris Do you think it’s feasible to use molecular simulation data as “low quality” biology data?

@mkoeris If we want similar breakthroughs in other fields, the formula is simple: find ways to collect cheap low-resolution data alongside a few high-resolution datapoints, then train models to do the upscaling. (2/2)

[1/3] Protein folding worked with ~250K PDB structures because the problem is degenerate. Most problems in nature aren't. Scaling laws hold broadly but the data doesn't exist yet. That's the infrastructure problem nobody is solving at scale. Evidence + links in thread 🧵

@yacineMTB My response when people ask why I’m using linear regression still

the more retarded it is the more likely it will work

@deboramarks 100% of designed binders successfully bind to their target. I don’t think that will be possible with only model improvements. We’ve essentially reached the limit of what can be done with the PDB data. Ideally there should be more work done on data ontologies for biology

@booleanbio 100% of what?

#AI for #proteindesign is a done deal so is it time to

@SimoneSyed Routine bloodwork seems ideal. It’s not constant monitoring but gives enough actionable feedback to actual improvement how you feel day to day

I'm not convinced that constant health metric monitoring is good for most humans

Does anyone else start feeling sick after spending too much time with the in vivo data? I went into computation because in vivo breaks my heart. Here I am again 💔

@parmita Just curious what it looks like to use your platform. We are building out our tox workflows right now to try to minimize adverse event risk.

Every AI CEO has promised to cure disease. They are LYING. Not because the AI is bad. Because the data doesn’t exist. The entire industry trains models on snapshots of dead cells in mice. That’s why 90% of drugs fail in humans. The FDA just told pharma to replace animal testing in 3-5 years. There is no replacement at scale. No one knows where to go. We are where you go. We built the instrument that captures what actually happens inside living human cells when you add a drug. Data that has never existed before. Data that will be what cures cancer and autoimmune and dementias. Every drug we screen trains a foundation model. The model gets smarter. The data compounds. the same biology that predicts toxicity reveals novel drug targets. entire classes of disease that are undruggable today because nobody could see them. We’re not an AI company making promises, we are making this problem actually solvable. We’re the data layer that makes everyone else’s promises real.

@yacineMTB Okay but can I do this w/biology data because that would be incredible

prediction: someone is going to get a coding AI like codex to automate turning existing steam video games into harnesses, come up with architecture to parallelize the games themselves in a manner that is conducive for RL training, and train an RL demigod model

@Gabogonzalez515 I imagine coffee wasn’t super hard to discover. You can boil all sorts of leaves and fruits to extract nutrition out of something that’s basically inedible. Food abundance is a modern invention so people were likely boiling all sorts of stuff to survive.

All the weird LLM hacks people have found to squeeze better performance out of their models (weird (system) prompts, repeating at beginning and end, no all caps, etc.) makes me feel the same way as like how the fuck did we figure out how to make coffee?

@parmita They can’t fathom that some people would rather save lives than program targeted ads.

Brother…I have my own lab. At 26. A sensor and machine that should be worth more than many Ferraris. We built it. Data that could actually save people. No. I don’t get FOMO when I look at any engineer at “FAANG.” I hope they enjoy the work they’re doing like I do mine.

@vesriram @iskander How many do you generally do as a panel of “alike” targets? I’ve been avoiding this because it can get somewhat expensive

@iskander Beyond what's been mentioned already - my workflow: DALI or Foldseek to find proteins with fold/domain similarity to target protein as first check screen. I also use CHO/293 lysates or sometimes baculovirus particle assay

Going to test 192 de novo binders against two target proteins + HSA & Hsp90 as negative controls. Are there some particularly sticky common proteins I should add to that control set?

@iskander Depends on your downstream use case, but I do think it’s worth it in general

@booleanbio You think it's worth doing both HSA and ovalbumin?

@kalomaze Applied AI is really meat+butter. Translating new AI models and innovations to real-world applications will always be around, and it’s extremely fulfilling to see how the ML/AI insights lead to real products!

data