Hanbin Lee

Scaling down protein language modeling with MSA Pairformer Recent efforts in protein language modeling have focused on scaling single-sequence models and their training data, requiring vast compute resources that limit accessibility. Although models that use ...

Excited to share work with
Zhidian Zhang, @milot.bsky.social, @martinsteinegger.bsky.social, and @sokrypton.org
biorxiv.org/content/10.1...
TLDR: We introduce MSA Pairformer, a 111M parameter protein language model that challenges the scaling paradigm in self-supervised protein language modeling🧵

Causal clarity in statistical software Imagine running a simple regression in any statistical software of choice—but this time, you only get a point estimate of the regression coefficient. There

Should statistical software that estimates causal effects also tell you the causal assumptions under which that estimate can be interpreted as causal?

I don't know but my PhD student Maurice Korf has some thoughts (and software) to get the conversation going:

academic.oup.com/ije/article/...

The History of the Panmictic Population Concept and Its Legacy in Contemporary Population Genetics ABSTRACT The panmictic population concept is at the heart of population, evolutionary and conservation genetics. However, in nature, true panmictic populations are vanishingly rare. As an idea conce...

onlinelibrary.wiley.com/doi/10.1111/...

Will read during comutes

The biggest blemish of mine is not being able to speak Japanese despite all that time watching animes.

commutes should count toward work hours

Always surprised by such ambitious titles, and then again by the use of clever data dug up from historical records.
I read this when it was a working paper but still...

One takeaway is that same-sex sexual behavior is not a special trait that needs explaining. Rather, it can follow from reasonable mating strategies under imperfect information. Indeed, attempting to mate *only* with the opposite sex is a derived trait that only arises under some conditions.

The 2026 Probabilistic Modeling in Genomics (ProbGen) meeting will be held at UC Berkeley, March 25-28, 2026. We have an amazing list of keynote speakers and session chairs:
probgen2026.github.io

Please help spread the news.

Looking south from the boardwalk around Lake Harriet. In view are 6 sailboats and a variety of brave folks about to board these vessels.

Nature is my psychiatrist. Walking down a tree lined cement path going toward the Mississippi River near Lake St in Minneapolis.

Looking east under the Lake-Marshall bridge. In view are the bridge pilings with graffiti and haze in the air.

Image of the front wheel of a moving bicycle casting a shadow of the wheel onto the gravel road.

Views from the urban hellscape they call Minneapolis. River, lake, gravel, he, they, she, radio, river otter, joy.

In a new Perspective article, Josh Morgan discusses how progress in cell biology is hindered by significance testing and the need for a shift to effect size estimation. rupress.org/jcb/article/...

#Technology #Reproducibility #CellCycle #CellDivision #Statistics

Finally a biblic reference in which I understand

Legend says the ancient Babylonians once tried to sequence and annotate God's own genome, and for their ambition and hubris they were forever cursed to have different annotation formats and standards so they could never do genomics with ease again.

New preprint: SBI with foundation models!
Tired of training or tuning your inference network, or waiting for your simulations to finish? Our method NPE-PF can help: It provides training-free simulation-based inference, achieving competitive performance with orders of magnitude fewer simulations! ⚡️

Super excited to see this out. What started as some math in a grant in 2020, to a student deciding to take this on in 2022, to published in 2025.

These things can take time and patience is key!

rip ozzy

If I were to write a program, I would just do Poisson regression with posthoc corrected variance like glmGamPoi because it's cheaper and doesn't lose power. Theoretically it should lose some amount of power relative to NB but I've never seem such a case in practice.

This is technically a quasi-likelihood test and not a negative binomial MLE. Nevertheless, for many reasons, people keeps mistaking it as a NB regression. In the literature, some people do Poisson reg. with posthoc correction and some do NB MLE under the same name causing confusion.

This is actually what glmGamPoi is doing. It fits a Poisson regression. It estimates the dispersion parameter afterwards. Unlike NB regression, the dispersion and the regression coefficients are not jointly estimated.

One thing that is overlooked in omics literature is that Poisson regression is correct not only for Poisson distributed data, but also for all sorts of count data as well as positive continuous outcomes (e.g Gamma distribution).
You only need to correct the standard error properly.

Flexible and efficient count-distribution and mixed-model methods for eQTL mapping with quasar Identifying genetic variants that affect gene expression, expression quantitative trait loci (eQTLs), is a major focus of modern genomics. Today, various methods exist for eQTL mapping, each using dif...

Very excited to share new work from my PhD on a new software package for eQTL mapping: quasar. The quasar software package is a C++ program designed to provide a flexible and efficient eQTL mapping. www.medrxiv.org/content/10.1...

Experimental evolution in an era of molecular manipulation - Nature Reviews Genetics In this Review, Ascensao and Desai discuss how methodological advances in genotype and phenotype manipulation are transforming experimental evolution approaches and providing new insights into the und...

Experimental evolution in an era of molecular manipulation

@natrevgenet.nature.com by @joaoascensao.bsky.social and @mmdesai.bsky.social

www.nature.com/articles/s41...

I also hate both of the publishers you've mentioned but most of the papers I read from post-90/00s always have a latex-typesetted preprint on the web, so the publisher doesn't matter. The problem mostly happens with older papers with both old notations and unfamiliar fonts pre-80s :(

Borzoi-informed fine mapping improves causal variant prioritization in complex trait GWAS Genome-wide association studies (GWAS) have identified thousands of trait-associated loci. Prioritizing causal variants within these loci is critical for characterizing trait biology. Statistical fine...

I'm excited to share work on a research direction my team has been advancing: connecting machine learning derived genetic variant embeddings to downstream tasks in human genetics. This work was led by the amazing Divyanshi Srivastava! www.biorxiv.org/content/10.1...

Experimental evolution in an era of molecular manipulation - Nature Reviews Genetics In this Review, Ascensao and Desai discuss how methodological advances in genotype and phenotype manipulation are transforming experimental evolution approaches and providing new insights into the und...

Experimental evolution in an era of molecular manipulation
#ExperimentalEvolution #evolution #evoSky

www.nature.com/articles/s41...

I guess this is a common theoretical econ problem with a caveat that I don't know much about ecoh.

Which Kind of Science Reform What hope is there for science reform, if we can't agree on what to reform? Right now, principles are more important than practices.

How can we reform science? I have some ideas. But I am not sure you’ll like them, because they don’t promise much. elevanth.org/blog/2025/07...

Is there a model that compares random grant assignments versus current merit-based award schemes? There must be a form of competition btw productivity loss due to sending money to less competent (whatever that means) ppl vs time wasted on writing grants/fiddling with administrative chores.

Am I the only person who sturbbonly seek for more recent references of the same content to avoid old papers with old typesetting?

okay I'm doing slim vibe coding

Currently reading Stigler’s History of Statistics again (and loving it). Anyone know of a similar book for 1900-1960 period. Efron & Hastie, is great but it’s a different kind of book. Apparwntly Lehmann wrote specifically about Neyman v Fisher but I’m looking for other options too #stats #statsky