Ethan Weinberger (@efweinberger) header
Ethan Weinberger (@efweinberger) profile picture

Machine learning for biology @UWCSE @suinleelab. @NSF GRFP fellow. Previously @yale

Seattle, WA
Joined November 2019
Photo by Ethan Weinberger (@efweinberger)Photo by Ethan Weinberger (@efweinberger)Photo by Ethan Weinberger (@efweinberger)Photo by Ethan Weinberger (@efweinberger)Photo by Ethan Weinberger (@efweinberger)
Ethan Weinberger retweeted
I think its becoming increasingly clear that trans perturbation prediction as the primary hill climbing objective is pushing the field further & further away from models that actually learn causal gene regulation. 1/
4
9
1
115
11,314
Ethan Weinberger retweeted
Single-cell atlases continue to map the diversity of cell states, but how can we use them to interpret non-coding genetic variants? Excited to share AlphaGenome 2, our model for predicting cell-type-specific regulatory activity and variant effects from DNA sequence. deepmind.google/blog/alphage…
12
82
7
291
39,981
Ethan Weinberger retweeted
Most students learn that DNA → RNA → protein. This sounds straightforward, but transcription (or that DNA → RNA bit) is surprisingly unpredictable, even in genetically identical cells. And, unfortunately, many tools commonly used in biology miss the way that gene expression actually works. For several decades, most biologists assumed that transcription followed a Poisson process. In other words, imagine that you are a gene within a cell moving through time. At each small step forward in time, you roll a die, and there's a fixed probability that this gene will get transcribed into mRNA. Most of the time, nothing happens; but sometimes you get a new mRNA. In a Poisson distribution, as time goes on, the probability that the cell has not made a transcript at all trends exponentially toward zero. Scientists believed this because they did not have methods to refute it. All the techniques we had to study gene expression were indirect. They might put a gene encoding green fluorescent protein into a cell, for example, let the cell transcribe and translate that gene, and then use a microscope to measure how brightly the cell glowed. They'd count up the fluorescent signal and try to work backward to infer how many mRNAs were responsible. But obviously these were proxies, and there's no correlation (in many genes) between protein levels and mRNA levels. So in 2005, Ido Golding and colleagues published what I think is one of the most beautiful and important papers in modern biology. They figured out how to watch and count mRNAs in individual living bacteria, and used this method to test whether transcription really followed a Poisson process. First, they engineered E. coli to make a green fluorescent protein fused to something called the MS2 coat protein. MS2 binds to a particular RNA sequence, called an MS2 binding site. So the bacteria were producing large amounts of this green fluorescent protein, which was floating around inside the cells. Second, they introduced another genetic construct, this time encoding a red fluorescent protein. But at the end of this gene, they added an enormous array of 96 MS2 binding sites. So what happens is that, when the cell transcribes the gene encoding the red fluorescent protein, the resulting mRNA also contains this long string of MS2 binding sites, which the green protein (fused to MS2) quickly gloms onto. And now, instead of having green fluorescence diffused throughout the cell, you get these very bright, green spots or "puncta" where the GFP molecules have accumulated on an mRNA. And because each mRNA recruits dozens of GFP molecules, the researchers could calibrate the brightness of these spots and estimate how many individual transcripts were present in a single cell. Importantly, they could also activate the red fluorescent gene using a small molecule called IPTG. (The gene was placed "downstream" from a so-called inducible promoter.) This meant they could add IPTG, start a stopwatch, take pictures over time, and then watch the bright green spots appear as the bacteria produced mRNAs. And they found that gene expression is sort of Poisson, but with a huge caveat. In a true Poisson process, the variance equals the mean. So if a gene produces an average of 10 mRNA molecules over some period of time, the variance in the number produced should also be 10. The transcripts appear at random times, sometimes close together, sometimes farther apart, but each event is independent. Instead, Golding and his colleagues found that the variance in mRNA abundance was about FOUR TIMES larger than a Poisson process would predict. This happens because transcription is BURSTY. A gene remains silent for a long time, then suddenly turns on and makes a bunch of mRNAs really quickly, and shuts off again. A cell might produce four transcripts over a few minutes and then make nothing for the next half hour. (The full model is this: When a gene is ON, transcription follows a Poisson process. But the gene can switch OFF at any moment. And because each ON period lasts a different amount of time, the number of transcripts produced in each burst follows a geometric distribution. In their experiments, genes were inactive for about 37 minutes on average, then active for about six minutes, producing an average of 2.2 transcripts per burst.) This study changed how we study biology. It's so important, because it influences the way we study RNAs in cells today. Depending on when you kill cells and extract their molecules, for example, some might have no transcripts from a particular gene, while others might have lots of them. That variation doesn't necessarily mean the cells are fundamentally different; some of them might have just had a recent burst of transcription, while others haven't. Or consider two genes that have exactly the same expression level. One might produce transcripts slowly and frequently, while the other produces them in large, infrequent bursts. If you just measure how many mRNAs are present, you can't tell how those transcripts were produced, or how the genes are regulated. I made a little interactive to show how this works and hope it is helpful.
16
119
12
744
40,669
Ethan Weinberger retweeted
Launching RNAFoundry.com – we manufacture personalized cancer vaccines.
334
1,297
264
6,186
833,505
Ethan Weinberger retweeted
Incredibly, incredibly excited to introduce SpliceAI2, the next generation of splicing prediction! 🧬🚀🧵
8
36
4
102
10,468
Ethan Weinberger retweeted
Great to the see the flurry of single gene knockdown Perturb-seq like atlases from cell-lines, mouse brain etc over the last few days. These are undoubtedly very valuable datasets. I just want to re-iterate a few other very important expt. design considerations 1/
4
77
11
336
37,646
Ethan Weinberger retweeted
One year to the day since defending my PhD thesis, I'm thrilled to share that our latest work is now published in @Nature! It's been an incredible journey exploring the DRT family, with many unexpected discoveries and never a dull moment.
1/5 Telomerase has long been linked to chromosome maintenance in eukaryotes. Now, we show that telomerase-like activity is also present in bacteria, shedding new light on its ancient origins. Excited to share that our work on DRT10 is now out in @Nature! nature.com/articles/s41586-0…
4
7
43
2,327
Ethan Weinberger retweeted
1/ Most genes and variants matter somewhere but look silent in any one cell model. Can we build conditions that reveal phenotypes? With @kanishkadey, we engineer cell states to expose hidden disease biology and trace how genetic risk reaches disease genes. biorxiv.org/content/10.64898…
6
113
11
407
56,687
Ethan Weinberger retweeted
After a long wait, tangermeme is out @naturemethods! tangermeme implements the primitives of genomic sequence-to-function modeling with efficient, tested, and composable functions. I use it every day in my work, and I hope you find it useful too. nature.com/articles/s41592-0…
7
85
3
299
19,805
Ethan Weinberger retweeted
A universal predictive model of the cell would dramatically accelerate science, allowing biologists to perform experiments digitally. Today we’re announcing a major partnership with the DOE and NIH. The goal of this partnership is to generate the data that will be needed to build an accurate model of the cell with artificial intelligence. Isomorphic, Google DeepMind, and Meta are joining us as founding partners in the Virtual Biology Initiative. This is the beginning of a large scale coordinated effort to map cellular biology to power the development of digital models of life. Other leading scientific institutions and consortia including the Allen Institute, Broad Institute, Gladstone Institutes, the Human Cell Atlas, the Human Protein Atlas, the Wellcome Sanger Institute, as well as NVIDIA, are working together as part of this international scientific project. Creating a predictive model of the cell is one of the most important challenges for the next decade of science. The insights that come from this could unlock a far greater understanding of disease, and open new paths for cures. We invite the worldwide scientific community and other funders to join us in this effort.
107
247
102
1,777
264,571
Ethan Weinberger retweeted
Would be grateful if the genomics community would read the recent ENCODE4 preprint and provide critical feedback. Link below. 1/
2
28
1
121
14,956
Ethan Weinberger retweeted
Very excited to announce ENCODE GRAMMAR (Genomic Regulatory Atlas of sequence Models, Motifs, Annotations & Rules): 3,865 experiment-specific deep learning model sets and sequence annotations for decoding human regulatory DNA. 1/
12
201
6
730
61,290
Ethan Weinberger retweeted
Amazing convergence of finding using orthogonal approaches - congrats Julia et al!
Replying to @juliabelk
Serendipitously, Nathan Zemke, Bing Ren, and @JoeEcker discovered the same phenomenon using a different approach: single cell methylation. Collectively, these data show that the replacement of yolk sac microglia by monocytes starts in middle age and is ~complete by age ~80. 5/ science.org/doi/10.1126/scie…
5
18
3,851
Ethan Weinberger retweeted
Your DNA is like the same instruction manual in every cell of your body. So why does one cell become a brain cell while another becomes a heart cell? The answer lies in epigenetics—the systems that tell cells which instructions to use and which to ignore. Salk scientists, along with collaborators, created the first body-wide atlas, profiling 86,689 cells across 16 human tissues, that maps both how DNA folds and how it's chemically tagged in the same individual cells. The result is an unprecedented look at how our cells are programmed and why disease risk may differ from one cell type to another. The freely available resource could accelerate research into conditions like schizophrenia, bipolar disorder, atrial fibrillation, and diabetes, and help train future AI tools for genetics. Read more: salk.edu/news-release/two-wa… Photo of Salk scientists from left to right: Jingtian Zhou, Jesse Dixon, and Joe Ecker. @ScienceMagazine @NIHgov #4DNConsortium @UCSanDiego @AAAS
10
3
27
7,265
Ethan Weinberger retweeted
New Genomics x AI blog post! @jmschreiber91 presents cherimoya an efficient seq2fun model for local regulatory function prediction! Great contribution. Love the thoroughness of hyperparam exploration and efficiency gains. Worth a read and studying repo! genomicsxai.github.io/blogs/…
41
3
150
27,485
Ethan Weinberger retweeted
Some of the accepted bioML papers are truly egregiously terrible. That's been the case at all the major ML conferences over the last many years. There was even a paper that won some kind of award last year at one of the conferences that was just chok full of fatal flaws.
a LOT of slop at icml it seems, we are hiring rn and it’s difficult to determine what’s actually substantial from a progress perspective in bio vs what’s just smth someone did over a weekend
10
11
2
191
64,462
Ethan Weinberger retweeted
[Reminder] MLCB 2026 submissions are open! Deadline: July 1 AOE Keynote speakers: @DrAnneCarpenter (Purdue), Christina Leslie (MSKCC), @ZhongingAlong (Princeton), and @BrennerMichael (Harvard/Google) Meeting: Nov 16–17 at NY Genome Center in NYC mlcb.org
1
15
1
40
5,199
Ethan Weinberger retweeted
7
15
1
150
13,119