iiinova — where your digital self keeps evolving

Reposted by Jacob Schreiber

Shenzhi Chen @shenzhichen1999.bsky.social · 25/08/2026

Excited to share our paper in @natgenet.nature.com on the de novo design of tissue-specific mammalian enhancers that function in vivo in mouse embryos. 🧬🐭 www.nature.com/articles/s41...

35826

Jacob Schreiber @jmschreiber91.bsky.social · 14/03/2026

There was concern it would not be helpful for identification purposes for the few people who do not already know me

010

Jacob Schreiber @jmschreiber91.bsky.social · 09/03/2026

The Programmable Genomics Lab at @umasschan.bsky.social just officially launched our site! We have a "simple" goal: to develop synthetic regulatory elements that target every cell type in every tissue, and control payload dosage/duration Check it out! programmable-genomics.github.io

programmable-genomics.github.io

Programmable Genomics Laboratory | Home

1218

Reposted by Jacob Schreiber

Alex Stark @alex-stark.bsky.social · 24/12/2025

Our preprint "Predictive design of tissue-specific mammalian enhancers that function in vivo in the mouse embryo" is on bioRxiv: www.biorxiv.org/content/10.6... . Amazing collaboration by @shenzhichen1999.bsky.social, Vincent Loubiere (@impvienna.bsky.social,@viennabiocenter.bsky.social),... (1/2)

biorxiv.org

Predictive design of tissue-specific mammalian enhancers that function in vivo in the mouse embryo

Enhancers control tissue-specific gene expression across metazoans. Although deep learning has enabled enhancer prediction and design in mammalian cell lines and invertebrate systems, it remains uncle...

210448

Jacob Schreiber @jmschreiber91.bsky.social · 10/12/2025

There's a ton of really cool stuff in the paper. Take a skim if you have time!

010

Jacob Schreiber @jmschreiber91.bsky.social · 10/12/2025

In collaboration with @alex-stark.bsky.social , we designed cell type-specific enhancers and experimentally verified them using STARR-seq. Strikingly, we found that Ledidi can even design enhancers with stronger regulatory activity than the strongest endogenous enhancers.

130

Jacob Schreiber @jmschreiber91.bsky.social · 10/12/2025

We designed regulatory DNA in dozens of settings and found that we could achieve our target objective with surprisingly few edits (biology alert). Because we focus on edits, our designs are less likely to overfit to the model or inadvertently alter properties that are not well captured by the model.

100

Jacob Schreiber @jmschreiber91.bsky.social · 10/12/2025

Ledidi is a DNA design method that designs *edits* to a template while explicitly minimizing the number of needed edits. This allows you to build upon informative starting material, which our genomes are full of, and simply edit in the final touches, similar to an Instagram filter or video touch-up.

101

Jacob Schreiber @jmschreiber91.bsky.social · 10/12/2025

After a huge amount of work w/ @alex-stark.bsky.social's group, a new version of our Ledidi preprint is now out! In an era of AI-designed proteins, the next leap will be controlling when, where, and how much of these proteins are expressed in living cells. www.biorxiv.org/content/10.1...

biorxiv.org

Programmatic design and editing of cis-regulatory elements

The development of modern genome editing and DNA synthesis has enabled researchers to edit DNA sequences with high precision but has left unsolved the problem of designing these edits. We introduce Le...

26026

Jacob Schreiber @jmschreiber91.bsky.social · 25/11/2025

The @impvienna.bsky.social is a unique place where talented researchers are doing amazing science. I'm glad I had an opportunity to spend some time there, and am looking forward to the next excuse to visit Vienna!

1120

Jacob Schreiber @jmschreiber91.bsky.social · 20/11/2025

That sounds too biologically important for me to be involved.

000

Jacob Schreiber @jmschreiber91.bsky.social · 20/11/2025

My first @umasschan.bsky.social/@impvienna.bsky.social affiliated paper is up! tomtom-lite is a re-implementation of tomtom targeting the ML age of genomics. Fast annotations ("what is this motif?") and simple large-scale discovery of motifs. Check it out! academic.oup.com/bioinformati...

academic.oup.com

Tomtom-lite: accelerating Tomtom enables large-scale and real-time motif similarity scoring

AbstractSummary. Pairwise sequence similarity is a core operation in genomic analysis, yet most attention has been given to sequences made up of discrete c

13514

Jacob Schreiber @jmschreiber91.bsky.social · 31/10/2025

Why is it a problem to translate C++ to changng standards when you can just use Agenic AI?

000

Reposted by Jacob Schreiber

Pinar Demetci @pinard.bsky.social · 29/10/2025

Tune in for a great MIA talk on ML for regulatory genomics by @jmschreiber91.bsky.social and Gregory Andrews now! 🧪 broad.io/mia (Will also be available online on our YouTube playlist later: www.youtube.com/watch?v=sMTO...)

052

Jacob Schreiber @jmschreiber91.bsky.social · 13/10/2025

they told the flight attendants to sit down for the second half of the flight (an hour) because it was entirely turbulence

200

Jacob Schreiber @jmschreiber91.bsky.social · 13/10/2025

had to land in a nor'eastern storm in boston and i'm ready to move back to the west coast forever

270

Reposted by Jacob Schreiber

Jacob Schreiber @jmschreiber91.bsky.social · 07/10/2025

Now that I'm settled in at @umasschan.bsky.social, I'm hiring at all levels: grad students, post-docs, and software engineers/bioinformaticians! The goal of my lab is to understand the regulatory role of every nucleotide in our genomes and how this changes across every cell in our bodies.

64326

Jacob Schreiber @jmschreiber91.bsky.social · 07/10/2025

I figure I can spend the time until I get tenure on answering the question, and then the time after tenure arguing about what "regulatory" means

170

Jacob Schreiber @jmschreiber91.bsky.social · 07/10/2025

If you're interested, please reach out with your CV and which topics you'd be interested in working on!

010

Jacob Schreiber @jmschreiber91.bsky.social · 07/10/2025

- Genomics Software Ecosystem: A major obstacle to our goal is the lack of simple+scalable software that everyone can use. Come build this with me. Training a lightweight deep learning model and using it for design/interpretability/VE prediction should be no more challenging than mapping reads.

160

Jacob Schreiber @jmschreiber91.bsky.social · 07/10/2025

- Foundation Models: As someone involved in ML, I am legally required to be working on this topic.

120

Jacob Schreiber @jmschreiber91.bsky.social · 07/10/2025

We have an array of ML-based projects for going after this, focusing on the following topics: - DNA Design ( 🧬 ) We have shown that Ledidi (www.biorxiv.org/content/10.1...) can precisely design DNA, and now it's time to push the boundaries in several directions w/ some very cool collaborations.

biorxiv.org

Programmatic design and editing of cis-regulatory elements

The development of modern genome editing tools has enabled researchers to make such edits with high precision but has left unsolved the problem of designing these edits. As a solution, we propose Ledi...

120

Jacob Schreiber @jmschreiber91.bsky.social · 07/10/2025

64326

Jacob Schreiber @jmschreiber91.bsky.social · 03/10/2025

It was suggested that the audience may not appreciate/understand :(

200

Jacob Schreiber @jmschreiber91.bsky.social · 30/09/2025

is it a good idea to wear a "join, or die!" hat to a big talk in europe? please say yes

130

Jacob Schreiber @jmschreiber91.bsky.social · 22/09/2025

the greatest productivity hack is having a grant deadline. there's so much other stuff you can do when you're supposed to be working on a grant.

0150

Jacob Schreiber @jmschreiber91.bsky.social · 19/09/2025

I was delighted to have the unexpected opportunity to give a keynote at MLCB 2025 in NYC last week. I used it to explain how I view deep learning models in genomics not as "uninterpretable black boxes" but as indispensable tools for understanding genomics + designing the next gen of synthetic DNA.

0121

Jacob Schreiber @jmschreiber91.bsky.social · 09/09/2025

for some reason i thought being a professor would involve more mentoring and research and less filling out disclosures concerning whether plants and seeds were used in my computational study

0101

Jacob Schreiber @jmschreiber91.bsky.social · 03/09/2025

stocking up the new apartment with essentials

040

Reposted by Jacob Schreiber

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

In the genomics community, we have focused pretty heavily on achieving state-of-the-art predictive performance. While undoubtedly important, how we *use* these models after training is potentially even more important. tangermeme v1.0.0 is out now. Hope you find it useful!

14414

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

For some reason, hitting "comment" on GitHub is significantly more responsive than a month ago and it freaks me out. Surely there are some important calculations that need to be done before letting my thoughts into the wild?

010

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

Thanks! Let me know if you want me to stop in virtually, we can try to figure out a time.

010

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

Hope you find tangermeme helpful in your work! Please reach out if you have any comments + questions.

000

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

Because everything is automatic, we can probe models. What motifs are driving model predictions? Calculate attributions, call + annotate seqlets, and count the annotations! BPNet is relying on MYC, whereas Beluga is relying on many more TFs. Easy comparison now.

110

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

Frequently, people manually annotate seqlets and draw bars or boxes around these high-attribution characters themselves. This is not really a problem, but it's just slow and does not scale genome-wide. In the above picture, everything is automatically done.

111

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

People *talk* about seqlets a lot but tangermeme is the first package for complete functionality. Here is a complete example of using tangermeme for attributions, seqlet calling + annotation, and plotting, to visualize what five models think of the same locus

130

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

Expanding past these implementations, tangermeme has a large focus on automatic seqlet calling and usage. Seqlets are short contiguous spans of high-attribution characters that usually correspond to the binding of a TF.

100

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

By considering attributions you can see how variants disrupt or change usage of motifs. Maybe you'll even find that a variant causes alternative binding by inducing a new motif or slightly changing competition! That would be challenging to see from the predictions alone.

110

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

Past simply re-implementing algorithms people use (in a convenient repo), tangermeme offers flexibility not usually offers in other implementations. As an example, instead of calculating variant effect as predictions before/after a substitution, why not look at attributions?

100

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

This care extends to each of our operations. For example, one-hot encoding the entirety of chr1 takes <2s on a single thread. This is significantly faster than other one-hot encoding methods out there, and is fast enough to enable real-time batch generation from FASTAs.

130

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

By focusing in this manner, we can "delve" deeply into these downstream algorithms. For instance, we found a bug in many DeepLIFT/SHAP implementations that will cause them to silently fail when you don't register your operations. Didn't know you needed to do that? Same!

120

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

This design choice is intentional. Because model components and training strategies are hugely variable and evolving quickly, I did not even want to touch those aspects. You define and train your model however you want, and then use tangermeme to do genomic discovery with it.

220

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

tangermeme is a toolkit that implements "everything-but-the-model" for genomic machine learning. This includes sequence manipulations, batched predictions, attributions, ablations, marginalizations, variant effect prediction, design, etc...

120

Jacob Schreiber @jmschreiber91.bsky.social · 27/08/2025

14414

Jacob Schreiber @jmschreiber91.bsky.social · 26/08/2025

It's about both. Conceptually, they are intertwined.

010

Jacob Schreiber @jmschreiber91.bsky.social · 26/08/2025

An excellent post about the receptive range of convolution models. "You might reasonably ask: "If I have 100 layers with W=1000W=1000, that's a theoretical receptive field of 100,000 tokens. Doesn't that matter?" The answer is no, and here's why:" guangxuanx.com/blog/stackin...

guangxuanx.com

Why Stacking Sliding Windows Can't See Very Far

Modern LLMs use sliding window attention for efficiency, but why can't stacking sliding windows see as far as theory suggests? A mathematical exploration of information dilution and the exponential ba...

150