
Dropping science like Galileo dropped the orange. VP DS & AI @AbaloneBio, ex-DS/ML @Atomwise @Zymergen, @grandroundsinc, PhD @UPennGCB
Bay Area, CA
Joined October 2012
- Tweets1.4K
- Following1.9K
- Followers398
- Likes5.2K
I phone @CA_EDD all days of the week x hours of operation to ask about PFL. Each time the call queue is full. Instead of arranging a callback as one would expect in 2024, the call is unceremoniously dropped. Each time. @GovPressOffice is this the best CA can do for new parents?
2
3
203
@BauerKahan would appreciate knowing your thoughts on how new parents should be supported by @CA_EDD
1
99
Multiple hypothesis testing strikes again: cureus.com/articles/196275-i…
1
125
This is a great way to encode inductive biases through pretraining tasks in biological datasets -- simulations! Important that they are non-trivial simulations. I have seen that simple tasks that amount to counting provide no benefit.
Compute Rosetta energies for protein variants and use them as pretraining targets to improve generalization on difficult protein property prediction tasks.
@romerolab1 @anthonygitter
1
236
I love my statistics peeps but while they are still gatekeeping semantics around _p-values_, ML folks are building the future. I guess it was inevitable but the gap is so much starker in 2024.
2
187
I'm late to the party because I use vim and just got @codeiumdev set up but it is clear to me that coding assistants are the (first) killer app of LLM's. I've used LLM's in other contexts but this is like it's reading my mind. I don't even have GPT4 with my free tier. Mind blow!
2
7
438
Sameer Soi retweeted
Scoop: Atomwise has hired a CMO w/ 25-plus years experience in the drug industry, as the AI biotech pivots from partnerships to pipeline
CEO Abe Heifets talked with me about the new focus, starting with TYK2, and why faster isn't necessarily better
endpts.com/atomwise-ceo-on-a…
Just updated my Zoom client for Ubuntu and see that both background blur and the whiteboard work. This really is the Year of Linux on the Desktop. Praise be!!
4
237
Sameer Soi retweeted
If this tweet gets 1k likes I'll waste my weekend benchmarking all the vector dbs to figure out what is real and what isn't. I'll throw `np.array` in the mix as well, and will use some "big data."
Sameer Soi retweeted
Omnigenic epistasis regulates complex trait heritability biorxiv.org/cgi/content/shor… #biorxiv_genetic
Sameer Soi retweeted
New pre-print alert 🚨
Genes and sites under adaptation at the phylogenetic scale also exhibit adaptation at the population-genetic scale. See 🧵below:
biorxiv.org/content/10.1101/…
Sameer Soi retweeted
Impact of phylogeny on structural contact inference from protein sequence data biorxiv.org/cgi/content/shor… #biorxiv_bioinfo
Sameer Soi retweeted
Excited to share our findings training GPN, a DNA language model for Arabidopsis thaliana, with @sanjitsbatra and @yun_s_song:
DNA language models are powerful predictors of non-coding variant effects, without the need for any labeled data.
doi.org/10.1101/2022.08.22.5…
1/n
Sameer Soi retweeted
@NucleateHQ's Activator is now OPEN for biotech company builders.
Accepting applications from academics worldwide. Bring your own idea or join a team.
apply.nucleate.xyz/
Sameer Soi retweeted
Another heuristic from @michelebusby:
"if your programmers are swearing at you in five years about your terrible code, that’s a win! You lasted five years and grew enough to hire judgmental programmers! Congratulations!"
(from this wonderful blog post: michelebusby.tumblr.com/post…)
Sameer Soi retweeted
Dear @ucla treating applicants and prospective and current employees with respect and dignity, and paying them a living wage, is the bare minimum. BARE MINIMUM.
Signed,
The Universe

















