←
Ruslan Rust4,664 posts
Ruslan Rust avatar

Ruslan Rust

@rust_ruslan

Asst Professor @USC with interest in the brain, BBB, stem cells, AI + what can go wrong after stroke & AD. Alum @ETH_en. Fixing science @ResearchHub

◷ Joined June 2019▧ 790 Photos and videos
Anshul Kundaje avatar
⟳ Ruslan Rust retweeted
Anshul Kundaje@anshulkundaje·8h

Scientific deliverables in the very near future (< 5 years) will diversify dramatically compared to the current monolithic definition of publication & cost/benefit for gate keeping peer review will continue to rise super exponentially making it effectively a complete farce.

My hot take is that it is still useful to have a scientific deliverable that corresponds to [a human spent a ~year thinking about this question, and the output was peer-reviewed], and that is the role academic papers should continue to play.

Ruslan Rust avatar
Ruslan Rust@rust_ruslan·8h

Brain border-associated macrophages rhythmically scavenge amyloid-beta and regulate vascular amyloid pathology: Immunity cell.com/immunity/fulltext/S…

Ruslan Rust avatar
Ruslan Rust@rust_ruslan·11h

This is really interesting. It’s not trivial to figure out which models work best for different bio tasks. Ideally, we’d have something like openrouter that tracks benchmarks and directs us to the best or most efficient model for each task. I also wonder how much performance changes when models have access to specialized tools and connectors. In my experience, that can make quite a difference, and benchmarks may not fully capture it (haven’t checked this one though!) really interesting work!

Claude was trained on protein sequences, at scale. We gave 20 language models the start of little-studied bacterial proteins and asked for the next amino acids. Only the latest Claude models predict them. Not GPT, not Gemini, not seven open-weight models.

Post attachment
Kexin Huang avatar
⟳ Ruslan Rust retweeted
Kexin Huang@KexinHuang5·15h

The most surprising part for me is that Fable 5.2 has massive gains over protein language model, for next amino acid prediction on less-studied proteins - where could these gains come from? Great finding led by @TuXinming @phylo_bio

Post attachment

Claude was trained on protein sequences, at scale. We gave 20 language models the start of little-studied bacterial proteins and asked for the next amino acids. Only the latest Claude models predict them. Not GPT, not Gemini, not seven open-weight models.

Post attachment
Ruslan Rust avatar
Ruslan Rust@rust_ruslan·14h

Highly recommended 👇

Applications are open for Astera Residencies, now through January 7, 2027! Up to $2M in funding, based out of our Emeryville hub. This round, we’re sharing a few specific project requests. Read more and apply here: astera.org/residency/ Reshare with folks who should see this!

Post attachment
Astera Institute avatar
⟳ Ruslan Rust retweeted
Astera Institute@AsteraInstitute·20h

We funded 10 grad students to publish their hidden science. The work fell into three groups: results with no home in a paper, datasets too valuable to sit on a server, and tools that help other scientists work. See the results ➡ asterainstitute.substack.com…

ResearchHub Foundation avatar
⟳ Ruslan Rust retweeted
ResearchHub Foundation@ResearchHubF·22h

Introducing SciCon 2026 speaker Dr. Jason Kelly. Co-founder and CEO of @Ginkgo, @jrkelly recently chaired the U.S. National Security Commission on Emerging Biotechnology. 🗓 December 10 📍 San Francisco

Post attachment
10x Genomics avatar
⟳ Ruslan Rust retweeted
10x Genomics@10xGenomics·Nov 16

Dr. Rahul Satija (NYGC) explains how VIPerturb-seq and Flex Apex make genome-wide CRISPR screens faster and more accessible—scaling to nearly 500K cells per lane with exceptional guide detection. See how: 10xgen.com/28m #SingleCell #CRISPR

Post attachment
OpenAI avatar
⟳ Ruslan Rust retweeted
OpenAI@OpenAI·Nov 16

Scientists reported nearly doubling their productivity with AI in our study of more than 1,800 participants in the ChatGPT for Academic Researchers program. Using privacy-preserving methods, we looked at how they use ChatGPT across their work—from developing ideas to analyzing data and writing code—and what it could mean for scientific discovery. openai.com/index/science-at-…

0:39
Laura Kida avatar
⟳ Ruslan Rust retweeted
Laura Kida@LauraCKida·Nov 16

About 90% of RNA molecules in our cells are tRNAs. To date, we have mostly been blind to their abundance and regulation in single-cell sequencing. Today, I’m excited to share a new manuscript from my MD/PhD work biorxiv.org/content/10.64898…

Ruslan Rust avatar
Ruslan Rust@rust_ruslan·Nov 16

I believe one of most impactful (and easiest) contributions frontier AI labs could make to science would be to run AI reviews of entire the open-access literature, including supplements, and make the reviews freely available to everyone. PMC has ~8.4M open-access articles. For this rough estimate, I used 3M papers as a starting point, focusing on original research and screening out low-quality stuff etc. Estimated model cost: $0.6–6M, depending on the model. Then keep it running for every new bioRxiv/medRxiv preprint. I believe this is a super small amount considering what we already spend on research. The US alone spent an estimated $993B on R&D in 2024. Even the upper estimate of $6M is less than 0.001% of that. + hopefully in the long run, preprints + public AI and human review will eventually replace journals.

Post attachment