iiinova — where your digital self keeps evolving

Reposted by Vitalii Kleshchevnikov, PhD

Nature Portfolio @natureportfolio.nature.com · 21/06/2026

The biological age of individual cell types can be evaluated using plasma proteomics, revealing diverse aging profiles across more than 40 cell types and links between the accelerated aging of specific cell types and disease, according to a paper in Nature Medicine. go.nature.com/4xHGvR2 #medsky 🧪

This is figure 1, which shows a framework for modeling cellular aging with plasma proteomics.

0329

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 21/06/2026

Smart frontier AI models in a good medical harness - that contains your full medical history and access to the same resources doctors have - will transform medicine in the next 1-5 years. It will reveal which doctors bring insight and which are simply gatekeepers who deliver worse outcomes than AI.

120

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 13/06/2026

Trying over the counter medication for a week to see if it helps resolve easily measurable symptoms is a very good idea. Ibuprofen, antihistamines are safe to take for a week. You might get evidence that your symptoms are driven by inflammation/mast cells that doctors will find harder to dismiss.

100

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 24/04/2026

Chatbots are a dead end. I rarely use them and whenever I use them they don’t quite do what I need. Next generation of coding agents is the future. More task tailored harnesses, smaller focused base LLM models.

110

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 02/04/2026

Do I know anyone who received Wellcome Early Career Award?

100

Reposted by Vitalii Kleshchevnikov, PhD

Craig Kaplan @triggerloop.bsky.social · 02/04/2026

I also think that one should always want to be in position of being the sharpest critic of one's own model, with as little emotion as possible. Our models are only useful if they serve the scientific pursuit and it is important that we challenge and test them. 9/

1106

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 31/03/2026

AI is getting really good at tasks with verifiable rewards namely coding. Computational biology feels like coding but it is not. Much of biology is about identifying and using symbolic causal representations of how things work, and hierarchical values guiding experiments, data, benchmarks.

173

Reposted by Vitalii Kleshchevnikov, PhD

Michael Sars Centre @msarscentre.bsky.social · 17/03/2026

We have some exciting guests in the @lionlchristiaen.bsky.social lab facility ✨These beautiful E. esculentus sea urchins collected near Bergen will help guest researcher Antoine Rémond from the @minclab.bsky.social better understand cleavage in echinoderms 🥚

Large pink sea urchins in a tub of water, together with some strands of green algae.

0218

Reposted by Vitalii Kleshchevnikov, PhD

Fabian Theis @fabiantheis.bsky.social · 17/03/2026

Excited to share new preprint with NVIDIA on GPU-accelerated single-cell analysis 🚀 rapids-singlecell brings native GPU support to scverse/AnnData: hours → seconds (1M cells: ~52m → ~25s), scaling to 100M cells. Huge thanks to Severin Dicks 🙌 📄 arxiv.org/abs/2603.02402

13814

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 17/03/2026

Before Claude Code: big picture planning is happening but the project is regularly stuck: implementation is a bottleneck - needs long uninterrupted chunks of time + optimal brain day. After Claude Code: how do I find the time for big picture planning when all these agents need to be instructed.

021

Reposted by Vitalii Kleshchevnikov, PhD

Kresten Lindorff-Larsen @lindorfflarsen.bsky.social · 08/03/2026

Review from Fia B. Larsen in @rhp-lab.bsky.social with everything you always wanted to know about proteasomal control of transcription factors (but were afraid to ask about) Proteasomal control of transcription factors: mechanisms, regulation and dysregulation. doi.org/10.1007/s000...

13011

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 07/03/2026

I don’t believe I am saying this but the AGI is finally here. Claude Code and Claude Cowork while not super intelligent, aren’t oracles that don’t make mistakes and aren’t very good at complex/ nuanced decisions and big picture engineering - they certainly feel like AGI. Incredibly helpful.

120

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 04/03/2026

We need to create more nuanced bullshit benchmark that collect questions from many labs about their area of expertise. It doesn't release the question but publicly shows how different models perform at the more nuanced questions, grouped by field and grouped by lab that submitted the questions.

010

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 04/03/2026

Pretty much everyone doing ML, AI, computational biology should be thinking about AI alignment. Getting the models to give useful insights about biology is one of the hardest parts and it’s most of what I do in these projects.

040

Reposted by Vitalii Kleshchevnikov, PhD

Danai Vagiaki @danaivagiaki.bsky.social · 08/02/2026

Delighted to present Latent Interaction Variational Inference (LIVI), a framework for trans-eQTL mapping at single-cell resolution that I developed during my PhD together with colleagues from @steglelab.bsky.social 1/n

12713

Reposted by Vitalii Kleshchevnikov, PhD

Stegle Lab @steglelab.bsky.social · 10/02/2026

Working on trans-eQTL mapping with population-scale scRNA-seq? Meet LIVI, our latest framework that enables efficient trans-eQTL mapping at single-cell resolution across cohorts of hundreds to thousands of donors. Preprint: doi.org/10.64898/202... Summary in the 🧵 below by @danaivagiaki.bsky.social

doi.org

0114

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 04/03/2026

AI alignment isn’t an abstract goal about preventing science fiction scenarios - it is central to figuring out how to make AI useful for all tasks. It’s quite impressive how the company that cares about alignment produced Claude Code and scored well on BullshitBench and low AI psychosis behaviours.

020

Reposted by Vitalii Kleshchevnikov, PhD

Gennady Gorin @goringennady.bsky.social · 03/03/2026

After years of work, the centerpiece of my PhD is published in @natmethods.nature.com! Read it to learn about the biophysical insights we can get from single-cell data! But first, I would like to talk a bit about RNA velocity and normalization. 1/

14015

Reposted by Vitalii Kleshchevnikov, PhD

Waggoner Lab @labwaggoner.bsky.social · 14/02/2026

@biorxivpreprint.bsky.social Hyperfunctional, gzmK-biased, glucose-responsive CD20+ cytotoxic T lymphocytes may contribute to the chronic low-grade inflammation characteristic of type 2 diabetes www.biorxiv.org/content/10.6...

083

Reposted by Vitalii Kleshchevnikov, PhD

VIB.AI @vibai.bsky.social · 13/02/2026

The @steinaerts.bsky.social lab is looking for a postdoctoral researcher to develop next-generation sequence-to-function models for glioblastoma, one of the most aggressive brain cancers. More info & how to apply 👉 vib.ai/en/opportunities#/job-descri…

0611

Reposted by Vitalii Kleshchevnikov, PhD

Waggoner Lab @labwaggoner.bsky.social · 02/01/2026

Generation of T cells with reduced off-target cross-reactivities by engineering co-signaling receptors CD5, CD8, or CD4(w/o changing TCR) www.nature.com/articles/s41... @natbiomedeng.nature.com @tcell.bsky.social @ox.ac.uk

051

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 24/12/2025

I recently tried long meditation sessions (1h+). Some insights are very hard to reach in just 10 minutes. Try starting with yoga nidra / body scan and switching to open awareness. Notice the difference between “looking with your mind’s eye” at where sensation happens and just sensation happening.

100

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 19/12/2025

Controversial take. Duplicated code is fine. It’s more important to wrap your analysis code into importable functions, even if each function duplicates some code. Taking a code that works and wrapping it into functions is often good enough. Code improvement activities expand to fill the time.

010

Reposted by Vitalii Kleshchevnikov, PhD

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 21/11/2025

Defining models and modelling tasks in ways that avoid “the model knows something that cells don’t” should be the guiding principle for making models.

021

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 17/12/2025

I don’t understand why people see in person work as a way to get people to work with each other instead of directly telling people to work with each other and providing them with tools to do so. If people want to work with each other they find a way.

200

Reposted by Vitalii Kleshchevnikov, PhD

Claudio Cantù @claudiocantu81.bsky.social · 13/12/2025

Can one map the genome-wide binding (1) and its protein partners (2) simultaneously from the same sample? Yes, one can. with CUT&ID ✂️🪪 Spearheaded — singlehandedly — by @annanordin.bsky.social No need of transgenesis, cloning and overexpression. Check it out, it's fast and its works.

617168

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 13/12/2025

There is a lot of content on how to start a startup. Much less or even no content on how to start a lab. Is that correct or did I miss anything? It would be quite helpful to do a series of interviews with people who helped recent PI start labs that explain what works, what doesn’t, what you do, etc

361

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 04/12/2025

From the list of things I DO recommend Spend a high fraction of time on strategy and planning compared to execution. It’s better to take the time to evolve ideas until they are good and execution takes minutes to days not weeks-month. Good ideas meaningfully address your most important problems.

110

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 04/12/2025

From list of things I really don’t recommend: Working like every week there is a paper submission deadline but for 6-18 months. I would abolish the deadlines in general in favour of evolving timeline projections. At specific dates, it only makes sense to report snapshots “this is how things look”.

020

Reposted by Vitalii Kleshchevnikov, PhD

Katherine Stiles @katherinestiles.org · 28/11/2025

· #AcademicSky · The European Research Council is launching €7m Plus Grants for ambitious, long‑term research — open to researchers at any career stage but limited to about 30 awards; current ERC grantees and applicants are ineligible, reports @clanicholson.bsky.social.

researchprofessionalnews.com

ERC’s new €7m Plus Grants open to researchers at any career stage - Research Professional News

European Research Council president describes scheme as part attempt to lure US talent

02520

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 21/11/2025

Defining models and modelling tasks in ways that avoid “the model knows something that cells don’t” should be the guiding principle for making models.

021

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 05/11/2025

How many hours per week do you talk to ChatGPT? I generally do 2-5h in total and several times per day - across coding, science, career, life, health. It’s becoming quite integral. Hard to believe I would say this given what I thought before.

100

Reposted by Vitalii Kleshchevnikov, PhD

James Briscoe @jamesbriscoe.bsky.social · 29/05/2025

Our latest "Dynamic Landscape Analysis of Cell Fate Decisions: Predictive Models of Neural Development From Single-Cell Data" A rigorous mathematical foundation for Waddington's landscape to study cell fate decision making Applied to ventral neural tube development www.biorxiv.org/content/10.1...

biorxiv.org

Dynamic Landscape Analysis of Cell Fate Decisions: Predictive Models of Neural Development From Single-Cell Data

Building a mechanistic understanding of cell fate decisions remains a fundamental goal of developmental biology, with implications for stem cell therapies, regenerative medicine and understanding dise...

513654

Reposted by Vitalii Kleshchevnikov, PhD

James Briscoe @jamesbriscoe.bsky.social · 29/10/2025

Good to see this study from Ribes & co in print: PAX3/PAX7 transcription factors control neural tube patterning by simultaneously repressing ventral fates via H3K27me3 deposition at silencers & activating dorsal fates as pioneer factors at enhancers @ribes journals.plos.org/plosbiology/...

journals.plos.org

Dual transcriptional activities of PAX3 and PAX7 spatially encode spinal cell fates through distinct gene networks

How do transcription factors, with pleotropic functions, generate organized cellular diversity in developing tissues? This study shows that PAX3 and PAX7 orchestrate spinal cord patterning by acting a...

0246

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 02/11/2025

Very exciting work. Reshaping understanding.

020

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 02/11/2025

I am wondering what people think about building an AI system to coordinate research. It would be an agent similar to ChatGPT 5. You can talk to it about your research with persistent memory across chats. All chats are used for training without opt-out option. 1/n

310

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 24/10/2025

Hot take. Excessive benchmarking demands on computational method papers slow down progress. Benchmarks in the paper need to show that your solution works well enough for your goals - not that it is the best solution ever proposed or how exactly it differs from all other existing solutions.

3101

Reposted by Vitalii Kleshchevnikov, PhD

Waggoner Lab @labwaggoner.bsky.social · 12/10/2025

@biorxivpreprint.bsky.social Scalable transcription factor mapping uncovers the regulatory dynamics of natural and synthetic transcription factors in human T cell states www.biorxiv.org/content/10.1...

062

Reposted by Vitalii Kleshchevnikov, PhD

zaira seferbekova 👾 @zaira-sef.bsky.social · 06/10/2025

🚨 new glioblastoma preprint alert! we present the first spatially resolved single cell atlas comparing radionecrotic changes (RN) and recurrent IDH-wildtype glioblastoma (GB) –– shedding light on a long-standing diagnostic challenge. 🧵 1/

1137

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 23/09/2025

How many papers are truly irreproducible and how many show a possible causal path which often isn’t the limiting factor and is conditional on important context (eg mutations, genetics, environment of the experiment)? Thinking about omnigenic model, “mechanistic bias” in medicine, personalised PRS.

010

Reposted by Vitalii Kleshchevnikov, PhD

Waggoner Lab @labwaggoner.bsky.social · 23/09/2025

Hepatic acetyl-CoA metabolism modulates neuroinflammation and depression susceptibility via acetate @cp-cellmetabolism.bsky.social www.cell.com/cell-metabol...

064

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 23/09/2025

Very impressive work and interesting modelling ideas 💡:

020

Reposted by Vitalii Kleshchevnikov, PhD

Yun S. Song @yun-s-song.bsky.social · 22/09/2025

By training GPN-Star on vertebrate, mammal, and primate alignments, we reveal task-dependent advantages of modeling deeper versus more recent evolution. These findings offer new biological insights and practical guidance for developing future gLMs and evolutionary models. (6/n)

142

Reposted by Vitalii Kleshchevnikov, PhD

Anshul Kundaje @anshulkundaje.bsky.social · 22/09/2025

This is truly an incredible breakthrough IMO. Really exemplifies what you get when deep domain expertise (popgen/evolution/disease genetics in this case) fuses with cleverly crafted ML. What u get r sleek, well thought out architectures that absolutely destroy the behemoths. Wow!! 1/

36414

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 16/09/2025

If you feel like you need noise canceling headphones 🎧 - you probably actually need to buy them right now. Don’t delay this decision and loose time trying to operate without them. Noise cancelling headphones are like your work email+slack - you may be able to work without them but it’s pretty hard.

130

Vitalii Kleshchevnikov, PhD @vitaliikl.bsky.social · 28/08/2025

Has anyone ever seen a house/flat in the UK/Cambridge that had an HRV or ERV system? If you can’t have windows open at night due to security concerns how are you supposed to ventilate? I tried closing windows and even with the fan on max CO2 levels never got below 850ppm (bad for sleep quality).

020

Reposted by Vitalii Kleshchevnikov, PhD

sebastiancachero.bsky.social @sebastiancachero.bsky.social · 21/08/2025

With unprecedented 38× coverage, our atlas resolves single neuron clusters and shows that separate waves of neurogenesis use different modes of molecular identity encoding: discrete in early born and continuous in late born. 3/8

191