Jeff Vierstra

Jeff Vierstra

@jeffvierstra.bsky.social

Senior Investigator @ Altius Institute for Biomedical Sciences. Research: High-resolution mapping of chromatin structure & function. Fun: Mountain shenanigans and skiing turns all year. Seattle, USA/Patagonia Chilena (🇺🇸🇨🇱). http://vierstra.org

405 followers401 following143 posts

Just wrapped up writing about what I have learned studying the noncoding mammalian genome over the course of the last 17 years. Should be on bioRxiv mid next week!

Been a while since I worked on the evolution of regulatory DNA. >12 years ago we described a phenomenon we coined "repurposing" where elements change cell activity patterns over time. ML has now enabled us to see the complex and subtle sequence changes underlying this.

Jeff Vierstra reposted

Here's the ad for the Genome Sciences faculty positions ap.washington.edu/ahr/position...

ap.washington.eduPosition Details - Academic Personnel & Faculty
Maitreya Dunham@maitreya.bsky.social

Exciting news: @uwgenome.bsky.social will be hiring for two tenure-track faculty positions this year, one open rank and one assistant professor. Both are “Genome Sciences” interpreted broadly so we welcome applicants working in a variety of fields and methods. Ad will be out shortly.

Jeff Vierstra reposted

You dont need advanced ML models to see base-resolution transcription factor occupancy and element-wise cis-regulatory logic. Just turn to a 40yo technology and basic stats. Still waiting for a structural biologist to help with 3d-modeling of composite regulatory DNA elements from DNase I profiles.

We are finally putting the final touches on an operationally complete mapping of regulatory DNA via DNase I in both human (>4,000 samples) and mouse. To interact with the samples we created this neat browser interface complete with a chatbot!

These data correspond to >1,000 individuals and we performed phased genotyping and allelic imbalance analysis with regulatory genomes (millions of SNVs). We have also uniformly processed and integrated >10,000 publicly available ATAC-seq datasets.

Exciting results! We developed a single, generalizable ML model that can predict chromatin accessibility across any arbitrary cell type using a sample intrinsic and portable embedding. Notably works on samples generated over a 15 yr time interval with different technologies & methodologies.

Using only training data from >4,000 human reference DNase I datasets we can also predict mouse DNase I. Also, it's readily tunable with the enormous amount of publicly available ATAC-seq data deposited to SRA.

We have also created a data portal complete with a coding chatbot to access the raw and processed data (>4,000 reference DNase I and >10,000 ATAC-seq samples). Coming soon to bioRxiv!

Surprised (but also not that surprised) that the AlphaGenome paper didn't officially cite any of the primary data used for training their model (see Fig. 1, thousands of datasets made with tremendous time and effort over >15yrs). What's up with that @nature.com ? www.nature.com/articles/s41...

nature.comAdvancing regulatory variant effect prediction with AlphaGenome - NatureAlphaGenome, a deep learning model that inputs 1-Mb DNA sequence to predict functional genomic tracks at single-base resolution across diverse modalities, outperforms existing models in variant effect...

With all the wild stuff going on in the States (and the world) I am going to escape reality for a while on a sailing trip to the end of the world around Cape Horn and the Beagle Channel (named after the HMS Beagle of Charles Darwin and Robert Fitzroy fame). Thinking this might be type 2 fun...

Great resource! I should mention (since it's not on the website) that all of the chromatin accessibility data (DNase I) was generated at the UW & Altius Institute over the course >15years. The proper references for these data are: www.nature.com/articles/nat... and www.nature.com/articles/s41....

nature.comThe accessible chromatin landscape of the human genome - NatureAn extensive map of human DNase I hypersensitive sites, markers of regulatory DNA, in 125 diverse cell and tissue types is described; integration of this information with other ENCODE-generated data s...
Open Targets@opentargets.org

The Open Targets Platform autumn release is out! 🍂 We have ingested the full dataset of over 13 million enhancer-gene regulatory interactions in the human genome across 1,458 DNase-seq experiments covering 369 cell types and tissues from the ENCODE-rE2G model blog.opentargets.org/open-targets...

Looks like a great couple of months of seminars! Come check out my talk on November 5th if you want to learn about our progress in mapping the nucleotide-resolved structure and function of cis-regulatory DNA elements across thousands of cell types and states.

FragileNucleosome@fnucleosome.bsky.social

We're super excited to announce the entire lineup for the Fall season of Fragile Nucleosome Seminars, starting on Sept 10th at 1200 EDT / 1600 UTC with @gracebower.bsky.social and @creminslab.bsky.social! register here for the entire series: us06web.zoom.us/webinar/regi...

Wild to see a thread about me. I think the broader topic (as Jason points out) is what does the future of preventive medicines look like for at risk gene carriers? I also hope this gives people some hope to those dealing with devastating and (previously) unactionable inherited genetic diseases.

Jason Mast@jasonmast.bsky.social

A 🧵: Last week, I wrote about Jeff Vierstra, who carries the gene for a devastating, rare form of ALS. The disease killed his mom and all three of her siblings, along with relatives dating back to the 1800s. Then, in 2020, two of his sisters began showing symptoms.

Does one sample (or even 10) suffice to define core cell type regulatory elements? NO! Because of both biological and technical variability you need to profile many (typically >15). The additional peaks are enriched for trait associated variants, so you miss a lot of possibly important signal.

Activity determining nucleotides on the BCL11A +58 enhancer according to a ML model built purely on DNase I data from thousands of cell types (this is just prediction for erythroid cells). Not bad w.r.t. functional data. The GATA1 site is the therapeutic target of Casgevy for SCD and B-thal.

For some reason I was re-reading the DEseq2 paper and was reminded of what a statistical masterpiece that method is. Every time I read the paper I seem to learn something new. Not too many papers achieve that bar (at least for me).

hotspot3: our chromatin accessibility peak caller is now a package – "pip install hotspot3" to try it out.

Jeff Vierstra@jeffvierstra.bsky.social

We have created a new DNase I- & ATAC-seq peak caller that uses an adaptive background model that controls for copy number variation & aneuploidy. It performs a per-nucleotide test (+FDR correction) and is very fast. Please try it out and give us feedback! github.com/vierstralab/...

You might know that my life mostly revolves around skiing. I am organizing a 25 day sail & ski trip to Antarctica in Dec. 2025 and have space for 1-2 more people. We leave from Ushuaia, AR on the Tierra del Fuego (early Dec.) DM me for details and pass this around if you know anyone interested!

We have created a new DNase I- & ATAC-seq peak caller that uses an adaptive background model that controls for copy number variation & aneuploidy. It performs a per-nucleotide test (+FDR correction) and is very fast. Please try it out and give us feedback! github.com/vierstralab/...

github.comGitHub - vierstralab/hotspot3: A chromatin accessibility peak caller with an adaptive background modelA chromatin accessibility peak caller with an adaptive background model - vierstralab/hotspot3
View full thread

We have applied this to ~4000 DNase I and >10,000 ATAC-seq (publicly available on SRA) datasets, and it seems to pass the "taste-test" as we call it. Maybe sometime in the future we will write this up -- though don't hold your breath 😂.

We have created a new DNase I- & ATAC-seq peak caller that uses an adaptive background model that controls for copy number variation & aneuploidy. It performs a per-nucleotide test (+FDR correction) and is very fast. Please try it out and give us feedback! github.com/vierstralab/...

github.comGitHub - vierstralab/hotspot3: A chromatin accessibility peak caller with an adaptive background modelA chromatin accessibility peak caller with an adaptive background model - vierstralab/hotspot3

This is cool and a blast from the past. Way back in grad school I spent like 10 months building a femtosecond laser to x-link TFs to DNA but could never get it to work.

Jakob Trendel@jtrend.bsky.social

Excited our paper is out in Cell @cp-cell.bsky.social! 🧬⚡ DNA photo-crosslinking proteomics in living cells 🎯 Pinpoints protein-DNA interactions to single amino acids 🌎 Globally quantifies DNA binding for >1800 proteins at a timescale of minutes 🔗 www.cell.com/cell/fulltex... 🧵