How to be as 'smart' as Auto-Regressive LLMs:
- memorize lots of problem statements together with recipes on how to solve them.
- to solve a new problem, retrieve the recipe whose problem statement superficially matches the new problem.
- apply the recipe blindly and declare

Introducing the Domainome... Site saturation mutagenesis of 500 human protein domains reveals the contribution of protein destabilization to genetic disease by Toni Beltran Centre for Genomic Regulation (CRG) Wellcome Sanger Institute BGI Genomics biorxiv.org/content/10.110…

Great GWAS paper diving deep into the contributions of germline variants to kidney cancer in 29k cases and >800k controls. I particularly love the discovery of a VHL 3' UTR variant, seen in ~10% of African and ~0.01% of European ancestries, that increase the risk of clear cell

UCLA Jewish student trying to attend class is forcibly prevented entry by 6 masked Pro-Hamas students while UCLA guard stands by and DOES NOTHING.

“I am a UCLA student…my class is over there. I want to use THAT entrance. Let me go in.”

#AntizionismIsAntisemitism

🎥

I always find it sadly ironic how quickly occupation protests manage to reinvent the concepts of policing, checkpoints, borders, and so on

We built CZ #CellxGene to provide the community with curated, interoperable single-cell data. Today, the platform has 85M+ cells & tools like Census that are being used to train new analysis models. Proud to see @nature highlight our work & collaborators nature.com/articles/d4158…

It pains me to see facile critiques of GWAS on here from our clinical/biostats friends while the many actually good reasons to be critical of GWAS get little attention. So here's a thread on what GWAS does, what critics get wrong, and where GWAS is genuinely still lacking. 🧵:

Very excited to share our new MPRA study on rare variants and cancer somatic mutations in the 3' UTR. Really proud of my former students Ting Fu and Kofi Amoah for spearheading this effort. Great collaboration with Sri Kosuri and team. nature.com/articles/s4146…

Missing cell types in single-cell references impact deconvolution of bulk data but are detectable ⁦Casey Greene⁩ biorxiv.org/content/10.110…

Arjun Raj People use AI without thinking basic science is 'intractable'. To me interpretable AI models like ChromBPNet are like the microscope - they have their biases and don't automatically produce understanding, but are a new way of looking that fundamentally helps understand biology.

Thrilled to release 🌟STaRK 🌟 - A large-scale LLM retrieval benchmark on semi-structured knowledge bases.

While LLMs excel at reasoning and semantic retrieval, they struggle with more complex tasks. Especially when real-world user queries require a combination of unstructured

You have perturbed TFs (for example disease vs healthy) and genome-wide H3K27ac data, what TF activity tools would infer the perturbed TFs that are more likely to be validated?

According to our results, the top 3 tools: RcisTarget, MEIRLOP and monaLisa

biorxiv.org/content/10.110…

The shift towards translational science and AI seems to be a consequence of thinking that basic science problems are either solved or intractable, respectively. Perhaps that is true, or perhaps we are mortgaging our future. I honestly don’t know which.

We're hiring! Looking for a Life Sciences Technician II to help us acquire RNA structural data on 1 billion diverse molecules: careersearch.stanford.edu/jobs/life-scie…

Why use Gnomix, our new local ancestry algorithm for big data?

1. Fastest, works w/ biobank scale data & whole genomes
2. Highest resolution, can find small, ancient segments
3. Most accurate
4. Performs phase error correction (& quickly)
5. Easy to use

github.com/AI-sandbox/gno…

Can we...just accept that we suck at this?

I want to tell you a story that has made me kinda hopeless about Twitter's ability to affect positive things happening, and it starts with this tweet from Hillary Clinton.

twitter.com/HillaryClinton…

I integrate and annotate 75 AML biopsies in part 2 of my 2024 single-cell guide. Custom label transfer, nuanced integration with hyperparameter tuning, avoiding pitfalls in analysis, etc. It's long, but one of my favorite videos so far. youtu.be/FqG_O12oWR4

AAV+Cas9 engineering leads to tandems/concatemers in 1/3 edited cells. Elegant and important paper with some practical solutions to the problem. Rehashing some prior work on this here 🧵 #Cas9 #ontargeteffects nature.com/articles/s4158…

Important update on metadata of H5N1 in cattle (and back to birds):

Thanks to the extraordinary detective skills of Flo Débarre, we are pleased to be able to share this table containing locations and dates for several H5N1 cases in cattle and birds:

github.com/andersen-lab/a…

apparently Google laid off their entire Python Foundations team, WTF!
( Aaron Gokaslan who is one of the pybind11 maintainers just informed me, asking what ways they can re-fund pybind11)
The team seems to have done substantial work that seems critical for Google internally as well.