James
224 posts
James
@basedsystems
cofounder @hoxbio | building integrated systems for discovery, health, and security
Katılım Aralık 2024
184 Takip Edilen631 Takipçiler
Scientists are getting their own “Waymo” moment — autonomous labs that take the wheel so researchers can focus on the big ideas.
@PharmaTechFocus interviewed @jrkelly on how our autonomous lab enables researchers to design, run, and iterate experiments at a scale and speed that manual science can't match.
Read the full Q&A by @Frankie_GD_PT:
hubs.la/Q04grlgW0
English
@anshulkundaje I know a guy who figured out how to generate the structures dataset without crystals. So, there will be much larger datasets in a bit of time
English
Firstly, AlphaFold is not a great example of the "Bitter lesson" IMO. It is trained on a relatively small labeled dataset & primarily transforms MSAs that encode structural constraints into structure with carefully crafted architectures with many inductive biases. 1/
English
"The Bitter Lesson has fully arrived in sequence biology and protein structure. Evo 2, AlphaFold 2 and 3, ProGen3, RFdiffusion".
This sentence has some issues IMO. 1/
Sylvain Gariel@SylvainGariel
English
@SashaGusevPosts will be a lot of this. probably need a way for centralized domain expertise to reach out to the datasets and pick at them live as like a service
English
We need a new clinical paradigm for AI algorithms that find incidental risk factors on a scan with AUC of 0.8 because these are (1) going to become more common and (2) don't have a clear path to commercial viability.
Derek Thompson@DKThomp
New newsletter: THIS IS HOW WE CURE PANCREATIC CANCER There are 3 broad reasons why pancreatic cancer is so hard to treat: 1. It often starts w a gene mutation that doctors thought was “undruggable” 2. It is largely “invisible” to the immune system by evading, and even turning down, T cell response 3. It’s asymptomatic until it’s too late—and radiologists can’t even see it on scans until it’s lethal We’re making progress along all three categories. Today’s post explains how—and how AI-assisted radiology could make a huge difference open.substack.com/pub/derekthomp…
English
James retweetledi
If you're building real things—as an engineer, entrepreneur, technician—consider having more children earlier than you thought.
Your children will have the irreplaceable inspiring experience of growing up with parents who do things and make things. More👇
English
in the future the large companies will have been started in group chats
Jon Erlichman@JonErlichman
First offices of 6 companies worth a combined $21 trillion.
English
James retweetledi
i think trump is working on some strange generalization of the collatz conjecture
unusual_whales@unusual_whales
Trump: These are tough questions. They say take a number, any number. I say ok, 99. Multiply it by 9. Divide it by 3. Add 4293. Divide by 2. Subtract 93. Divide by 9. What is your answer? I got it right. It was actually more complex than that.
English
James retweetledi
Update on Erdős Problem 1196:
In joint work, we refined and adapted the proof method from GPT-5.4 Pro to give proofs of several additional problems. This includes another 60 year old conjecture by Erdős, Sárközy, and Szemerédi.
A proof is valued not just by the problem it solves, but by what new avenues it opens up. This is perhaps one of the first examples of an AI-generated proof having downstream impacts, which we are still exploring.
We are announcing the result today at the Future of Mathematics Symposium (see links below)
English
@bazarovturg also, we rip apart how the standard software works because we need to optimize for correctness in our services. often you dig in and find missing knobs or broken / complex code
English
@bazarovturg we use it to run various internal services. better to just write a new, efficient one that integrates with the rest of our stuff
English
We wrote our own hyper-efficient version of the STAR aligner.
Typically it costs around $35 in cloud compute to process 30k cells with Cell Ranger. Ours costs so little that we don't charge for it. Also ours works with nanopore.
If you prefer pseudoalignments, then you can run it in pseudoalignment mode -- obviously we don't charge for that too.
All results instantly integrated with our other tools. As soon as a run finishes you can open alignments or expression matrices in our viewers without moving anything.
Let me know if you want to try it!
English
@basedsystems @chapmanb Does it work with other tech like Pip-seq from illumina? There Barcode structure is complicated as hell
English
In entrepreneurship we call this “effectual reasoning”
Doc Strangelove@DocStrangelove2
Whoever said a shopping cart can't be used as shooting platform was in fact wrong.
English
@basedsystems Weird comparison; normally one would say it's X times faster. $30-40 is a negligible amount in mol bio exp.
English
@SashaGusevPosts @parmita yea, problem is truth set is determined by the … tools themselves. phil discovered this when we wrote our snv caller
English
@basedsystems @parmita Yeah huge chicken and egg problem that we can often only efficiently benchmark the targets that we can effectively call/generate with existing tools.
English
you should ask my cofounder about how difficult it is to get good yield from cDNA and nanopore 😂😂😂
really this is a problem where chemistry and software needs to be co-designed
James@basedsystems
We wrote our own hyper-efficient version of the STAR aligner. Typically it costs around $35 in cloud compute to process 30k cells with Cell Ranger. Ours costs so little that we don't charge for it. Also ours works with nanopore. If you prefer pseudoalignments, then you can run it in pseudoalignment mode -- obviously we don't charge for that too. All results instantly integrated with our other tools. As soon as a run finishes you can open alignments or expression matrices in our viewers without moving anything. Let me know if you want to try it!
English