new post: LLMs have been widely successful in various domains, but how do they perform with proteins?
i wrote about protein language models, how transformer models have been applied to protein data, and the discoveries we made along the way.
i don’t like to participate in dating discourse so this is probably the only thing ill say about dating. but i find that posting private conversations and profiles from dating apps is such an incredible intrusion of privacy. we should all seriously stop doing it
being a statistician is fun cause you can tell people that the data they worked so hard to collect isn’t enough or won’t answer the questions they want answers to and move on while everybody hates you
✨introducing curie v1.0, a knowledge discovery tool to help you explore and learn any new topic in STEM.
dive into complex topics with interactive mind maps and get useful resources for each part.
if you're interested in trying it out, reply or DM to get access.
@_buildspace
i skimmed this paper. some quick notes:
- they curated the most extensive CRISPR operons to-date which they're calling CRISPR Cas-Atlas.
- they fine-tuned ProGen2 (a LM built to generate functional proteins) with this dataset to generate novel CRISPR-Cas proteins
- the new
One of the first projects we took on at
@ProfluentBio
was designing novel gene editing proteins with language models. This grew into an initiative called OpenCRISPR, and today I’m excited to share that work (including the sequences we designed)!
reading
@lakens
paper on p-values and one question that stuck with me is.. why don't we have statistical software that provides a misinterpretation-free verbal interpretation of the test instead of just spitting out p-values? seems like just doing that would solve many problems
GPT4 can recalculate p-values based on a given test statistic and degrees of freedom and check whether it matches the reported p-values.
huge for peer review/literature review and the replication crisis more broadly
our work in diagnosing and staging ovarian cancer with blood samples, led by chirag dhar is now published!
- we find a panel of glycoproteomic biomarkers that can distinguish between benign and malignant ovarian cancer with high accuracy and
- we
It's never been a better time to Build in Biotech 🧬
I'm really excited to be a judge at next week's hackathon brought to you by
@nvidia
@pebble_bed
and
@CromaticHQ
Interesting opportunity for the
@bitsinbio
community 👀
@3blue1brown
it shows up in rivers! more precisely, in the bendiness of it. turns out the average sinuosity ratio of rivers around the world approaches pi!!1!!1
the human genome project was completed in 2003 and it was the biological equivalent of putting a man on the moon-- predicted to change everything we know about disease and biology. but that was 2 decades ago, what's happened since? I wrote about it here:
We spent the last ~15 months building 🕹️ – our take on a modern, interactive React course.
Today we're happy to announce that it's live and available for everyone.
Here's why we're excited about it and why it took so long to create.
i'm new to bayesian inference and i've been learning a ton! i wanted to write this introductory post so (hopefully) others who are also just getting started can find it helpful.
its kind of an interesting feeling when someone moves slightly to avoid walking into you. it proves you exist. your mere existence made someone alter their path, can’t do that when you’re dead
being sick is kinda like a meditation retreat. the only thing you can think about is how weak your body is, so you get to forget about the rest of the world for a few days. pain becomes a focus, a meditation. and through it all, there’s assurance that you’ll get better soon
i've come to view human learning primarily under the lens of hiding complexity. i think that the only way to navigate the world is to view it on many different layers of abstraction, such that each layer describes reality in a way that hides 99%+ of what’s really going on.
big thanks to
@swyx
and
@Benghamine
for an incredible conference! here's some of my takeaways:
-2023 was the year of chatbots but they're not the future. how can we best design interfaces that handle model errors, understand and adapt to human behavior, and augment creativity?
@m_ashcroft
just today i learned that he was the first to propose the concept of a "universal constructor", a self-replicating machine that can be programmed to construct any other machine in 1940s. and sure enough, a decade later, DNA was discovered.
@nikillinit
I used to think this too but a lot of people simply don’t have the personality for it.. like the people that care more about stability than impact
it's absolutely tragic a lot of biotech startups in India with very promising data and bright founders are going under because the people with the capital just don't get how biotech funding works.
@salimfurth
@tylercowen
Bayes was just toying with the idea in 1760s through examples and it wasn’t formalized as a statistical concept until the 20th century
@rossabaker
also I think most people try to learn git by just trying it but it’s actually a lot easier if you spend some time trying to understand it before using it
@tszzl
this is an outcome of the immigration system, not an inherent immigrant trait. if you’re first gen, you were probably on visa for a long time and it is v limiting on how much money u can make (for eg, cant have multiple source of income) and hence the mindset
i am once again inviting y’all to another biotech meetup in NYC next friday(06/24)! this time we got a sponsor (thanks
@pillar_vc
) and a rooftop bar with an iconic view. come hang with us!!
the gundo boys aren’t the only ones tinkering with bits and atoms in town.
i’m co-hosting a
@bitsinbio
happy hour in LA on march 22nd to get people in biotech together. first of many events for the yr.
sign up below!
i don't know if this is gonna work but..
📢 i'm on the job market!
👩💻looking for data scientist/ statistician roles in healthcare
⌨️mostly R, sometimes Python
🎓MS biostat + BS biotech
🇺🇸require visa sponsorship
please RT ❤️
funny when you start to like someone, you might like them cause they're funny or smart or whatever. but when you get to a point to loving them, everything gets reversed. and it's not that you love them for those qualities, you love those qualities cause you love them
most conversations about accelerating science with AI is about accelerating scientific discovery which is cool and good but there’s tremendous alpha in accelerating scientific reporting and publishing that’s rarely ever talked about
the thing i love about imdb top 100 is that it never left the 2010 internet era. christopher nolan is pretty epic and we’re still reeling from al pacino’s performance in the godfather
a bit of a late update but i've joined
@nicktatonetti
's lab at cedars sinai and it's been an absolute blast! we're working on projects ranging from advancing SOTA research in precision pharmacology, translational bioinformatics, and implementing LLMs into clinical workflows.
as someone who's building an app in react with zero skills or experience in web dev, chatgpt has done more to derail than help me. it's been said before but it really cannot take you 0 to 1 on it's own. believe me, i tried
LOVE USA and HATE the immigration system??? im starting a signal group for the O1/H1B/J1 hustlers to share resources/tips/support. if interested, DM me your phone # and I’ll add you.
this is probably one of my favorite papers that combines EMR data, LLMs, and causal inference.
they provide an end-to-end framework to automatically simulate clinical trials from real-world data, including unstructured notes. biomedical language models
i am a huge proponent of incorporating knowledge graphs in ML to improve performance and this work on using KGs on smaller datasets where p>>n is very exciting!
Can deep learning work on small data with far more features than samples? We present PLATO: a method that achieves the state-of-the-art on such datasets by using prior domain information! 🧵
Published in
#NeurIPS2023
with
@ren_hongyu
@kexinhuang5
@jure
we're having our first
@bitsinbio
meetup in New York on 02/15. if you're interested in software x biotech, you should definitely stop by!
🧬where: The Brooklyneer
🧬when: 6:30 ET
see link below for more details and DM me if you have any questions.
best vibes at the AI x UX event last night! i demo-ed my project and got a lot of thoughtful feedback. thanks
@swyx
@thesephist
@Mappletons
for organizing!
if sama and greg start an AI company, do you think they’ll train another language model or just fine tune llama or something to catch up to where they were
imo one of the most crucial skills in data science that's rarely discussed is building an intuition for the magnitude of things. brian kernighan calls it numeric triage-- it's the ability to discern if a number thrown at you is much too high, or way too low or plausible