Xi Fu
@fuxialexander.bsky.social
110 followers 450 following 28 posts
Transcription regulation; deep learning; (bad) developer
Posts Media Videos Starter Packs
fuxialexander.bsky.social
My friend tried to ask what are some well known gene in Chr10/etc and guess what…
Reposted by Xi Fu
chaohou.bsky.social
We have updated our protein lanuage model trained on structure dynamics. Our new models show significant better zero-shot performance on mutation effects of designed and viral proteins compared to ESM2. check the new preprint here: www.biorxiv.org/content/10.1...
fuxialexander.bsky.social
much better with GPT
Reposted by Xi Fu
zhoujt.bsky.social
1/10 Excited to share our latest - the first whole-body map of both DNA methylation and 3D genome at single-cell resolution.
fuxialexander.bsky.social
Biorxiv seems to be really slow nowadays. Is it just me? Curious whether it's due to some infra change or there are some AI Agents crawling the data...
fuxialexander.bsky.social
Looks like a Tesla logo 😈 (I'm sorry...)
Reposted by Xi Fu
zuckermanbrain.bsky.social
For decades, government funding “has positioned the United States as a global leader” in science, says scientist Tom Maniatis of @zuckermanbrain.bsky.social and the New York Genome Center. He highlights how a new #NIH policy cutting money for research “jeopardizes” this, in Cell tinyurl.com/ubw6uphe
We must act swiftly and decisively to safeguard the future of science in the US - Tom Maniatis, PhD
Reposted by Xi Fu
tuuliel.bsky.social
Can someone send this to the NIH Director nominee who said yesterday under oath that he doesn’t know where the indirects go.
Reposted by Xi Fu
cbibberson.bsky.social
We are crowd sourcing reductions in graduate admissions and hiring freezes across biomedical research and higher ed in response to pauses in NIH funding and EO’s. If you have information if you could add to this spreadsheet, it would be greatly appreciated!: docs.google.com/spreadsheets...
Graduate Reductions Across Biomedical Sciences (2025)
docs.google.com
Reposted by Xi Fu
tuuliel.bsky.social
This is very cool work (where I was fortunate to play a small part), providing creative and crucial solutions for secure and federated eQTL mapping. Bigger functional genetic studies with less administrative and legal hassle! 💪
fuxialexander.bsky.social
Sorry for missing the final part of the first sentence:
...explained by known motifs of its own (increased overall affinity) or other TF (thus representing TF-TF interaction/cooperation), *or some other unknown factors*
fuxialexander.bsky.social
can't fit it in tweet and sorry for the formatting...
fuxialexander.bsky.social
The rest of the GC tracks might be more related to the 1D search/hopping hypothesis of TF binding as they seems to be less similar to the motif (unless those are for KLF/SP)
fuxialexander.bsky.social
it seems in this case, part of the context (~52-57) is a reverse complement of CAGC, resembles part of the less informative flanking (40-45) of the core GATAA/TTATC motif. I guess I would consider this still a “known motif” as probably it’s from some weaker DBD-DNA interaction
fuxialexander.bsky.social
How many information on the ledidi edit matrix can be explained by known motifs (of the TF itself or other TF)? Kind of confused right now on the role of context region in determine binding activities…
Reposted by Xi Fu
rkoszul.bsky.social
Deep learning models (@chromozz.bsky.social) trained only on yeast chromosomes predict nucleosome positioning, RNA Poll II and cohesin tracks along foreign DNA, based on the sequence alone. This implies that the behavior of any DNA in a host cell follows deterministic sequence-based rules.
CNN predict nucleosome positions, and RNA Pol II and cohesin deposition
Reposted by Xi Fu
pkoo562.bsky.social
[SAVE THE DATE] MLCB 2025 is happening Sept 10-11 at the NY Genome Center in NYC!

Attend the premier conference at the intersection of ML & Bio, share your research and make lasting connections!

Submission deadline: June 1
More details: mlcb.github.io

Help spread the word—please RT! #MLCB2025
fuxialexander.bsky.social
One probably one can also do that with MSA or even dig into bowtie-index file though 😂
fuxialexander.bsky.social
I believe one usage of the gLMs that truly align with it's training goal is discovery of DNA-binding proteins: train on many species, ideally jointly with (compressed) protein embeddings; identify overrepresented kmers specifically in certain clade; followed by massive oligomer-pulldown screening
fuxialexander.bsky.social
This looks like a python/scipy numerical lower bond. Sometimes I see R-based methods give much more floating digits. Has been curious what's the implementation difference.