Alex Predeus 🇺🇦

The 20th century saw the style of scientific figures and writing go from whimsical to minimalist, the 21st century from minimalist to quasi-brutalist. Given that AI can mimic this new affect ~perfectly, perhaps we will see a return to whimsy to signify our humanity?

@komposterov Написал в личку

linus was the original vibe coder before it was cool. dude just posts an angry email on the mailing list describing what he wants and thousands of engineers worldwide implement it for free. zero tokens consumed, zero API costs, infinite context window (30+ years of kernel knowledge). openai and anthropic are basically trying to replicate what linus has been doing with human contributors since 1991 except linus's agents dont hallucinate and they work for free @sahill_og

@komposterov Пока-пока-покачивая, я надеюсь!

The empty drops you threw out in your single-cell RNA sequencing analysis might be hiding mysterious things 👀 check out our bioRxiv preprint! The empty drops do not contain cells. Yet we can still use them to learn interesting things about biology and technology. 1/

@arjunrajlab Same with "functional validation", right? If you absolutely must find the difference, you will keep trying. Something's gotta give.

There is an interesting multiple hypothesis testing problem coming where you ask whether it's statistically valid to have a cool analysis if you run 100 (1000? 10,000?) analyses on a dataset and pick the best one.

How mandatory registration influenced clinical trial results

Spatial transcriptomics needs to look beyond just mRNAs. Great to see this work from @ntekasi, Lena Takayasu, @IwijnDeVlaminck, and me in a past life out today!

Fallout continues from one of the most appalling cases of scientific misconduct I've come across: journal retracts 138(!) case studies as fictional. retractionwatch.com/2026/03/03/can… New Yorker article that set off the current furor: newyorker.com/magazine/2026/…

Why are all of the websites in embl.de domain work so poorly? I'm trying to use metaTraits and eggnog-mapper, and they are down most of the time. It's been an issue for months @embl @EMBLHeidelberg

A quite amazing first-in-human trial in spina bifida Placenta-derived stem cells were applied onto the fetus’ exposed spinal cord in the womb The stem cells help to protect and restore the developing nerves before birth, which the current approach (surgery) doesn't solve The key question is whether this means that long-term function is better — that's now being assessed in a bigger trial!

This AI ad is wild 😂 twitter.com/brahma_4u/stat…

A century ago, “vitamin hunters” discovered micronutrients. Today, vitamins are taken adhoc. We revisited this with modern genetics: CRISPR screens -> new NAXD disease mouse -> over 40× lifespan increase w/ vitamin B3. Huge credit to Ankur & Skyler! tinyurl.com/32k74629 🧵👇

The NHS-Galleri trial results came out last Friday. 142,000 participants. 3 years. The most ambitious cancer screening trial ever run. The headlines: primary endpoint missed. That's a fact, and it matters. But I believe the headlines are missing the deeper story in these data – one that matters enormously to patients and to the future of cancer screening. Here's what the NHS-Galleri Test study showed: ✅ Stage IV cancer diagnoses – metastatic, typically incurable disease – declined substantially in the screened group, with >20% reductions across the 12 deadliest cancer types ✅ More cancers were found at Stage I & II, when treatment can be curative, in cancers that are typically diagnosed late (pancreatic, ovarian, liver, lung) ✅ 4x higher overall cancer detection rate when Galleri was added to standard screening (i.e., colonoscopy & mammography) ✅ Fewer cancers detected through emergency presentation – the worst possible way to learn you have cancer, and the most lethal and most costly ✅ No serious safety concerns. Test performance consistent with prior studies (with very high 99.5% specificity and very low false positive rate) So why did the 'primary endpoint' miss? See article for more detail & perspective on the study design, and the 'Stage Migration paradox' that masks enormous progress and clinical impact for patients.

Inspired by @JswLab, we generated a mini Genome-wide Perturb-seq, using just two 10x lanes (!). Far too much data for one tweet (or one Figure), but it works beautifully. The ability to assess the molecular function of every gene in an afternoon is mind-boggling

Comprehensive assessment of the possibility to build accurate predictors of pathogen exposure with T-cell receptor repertoires of 1000+ donors with known COVID19 status. Using both alpha and beta chains, HLA haplotypes, fixing batch effects and validating using two large independent cohorts link.springer.com/article/10.118…

@LM10_G8AT @nocontextfooty last season was far worse than this one, this year at least he seems in control of the team + bought some great players

@nocontextfooty probably the worst City side since Pep's arrived has peak Arsenal sweating

@Lefxpns @tangming2005 it's an R vs. Python difference (Seurat/Scanpy), and I'm not kidding

1/ If you're in bioinformatics, you're staring at matrices all day. RNA-seq? Gene x sample. scRNA-seq? Gene x cell. Everything is a matrix. But I never learned how to think in matrices. And I regret it.

New in Nature Neuroscience: we identify a previously unrecognized barrier at the choroid plexus–brain-CSF interface. nature.com/articles/s4159… 🧵👇

@slavov_n Large-scale and/or complex rearrangements, deletions and insertions are very hard to find with short reads. That's probably the lion's share of that 181.

No driver mutations could be identified in 181 tumors. The authors do not know why. Perhaps, it's an algorithmic failure. Perhaps, it reflects non-genetic contributions to carcinogenesis. What do you think ?