Very proud of two new preprints from the lab: 1) CREsted: to train sequence-to-function deep learning models on scATAC-seq atlases, and use them to decipher enhancer logic and design synthetic enhancers. This has been a wonderful lab-wide collaborative effort. www.biorxiv.org/content/10.1... - ThreadSky

steinaerts.bsky.social • 47 days ago

Very proud of two new preprints from the lab:
1) CREsted: to train sequence-to-function deep learning models on scATAC-seq atlases, and use them to decipher enhancer logic and design synthetic enhancers. This has been a wonderful lab-wide collaborative effort. https://www.biorxiv.org/content/10.1101/2025.04.02.646812v1

Comments

mads100tist.bsky.social•47 days ago

Congratulations, Stein and all!

anshulkundaje.bsky.social•45 days ago

Great work Stein. Just reading through it. U have a sentence in the intro regarding ChromBPNet "Footprinting modeling does require a large amount of data (large number of cells with high sequencing depths ideally over 100M reads)." This is incorrect as we explicitly show in the ChromBPNet paper.

steinaerts.bsky.social•45 days ago

Thanks Anshul, we'll revisit in the next update. With small clusters of 100-200 cells from scATAC-seq data, footprinting modeling works too? More feedback very welcome :-)

anshulkundaje.bsky.social•45 days ago

We have gone as low as 5M reads and we can still recover a lot. Its in the paper.

steinaerts.bsky.social•47 days ago

And a detailed thread for the CREsted paper: https://bsky.app/profile/niklaskemp.bsky.social/post/3llw4igqnsc23

steinaerts.bsky.social•47 days ago

with a detailed thread for the HyDrop paper:
https://bsky.app/profile/hannahdckmnkn.bsky.social/post/3lly22zma7k24

steinaerts.bsky.social•47 days ago

We investigate new CREsted models for the mouse cortex, PBMC, glioblastoma, and zebrafish development; and we validated synthetic enhancers for the heart, muscle, and endothelial cells in vivo in zebrafish. We furthermore fine-tune Borzoi to better capture cell type specific enhancer logic.

steinaerts.bsky.social•47 days ago

2) HyDrop-v2: with a new bead design it provides scalable and cost-effective generation of scATAC-seq atlases. With HyDrop atlases of the fly embryo and mouse cortex we show that CREsted models trained on HyDrop data are equivalent to models trained 10x atlases. https://www.biorxiv.org/content/10.1101/2025.04.02.646792v1

steinaerts.bsky.social•47 days ago

Thanks to the best microfluidics expert Suresh Poovathingal @s-poovathingal.bsky.social & congrats to all authors @niklaskemp.bsky.social @seppedewinter.bsky.social @hannahdckmnkn.bsky.social @erencaneksi.bsky.social @casblaauw.bsky.social Vasilis @lukasmahieu.bsky.social @samdieltiens.bsky.social

Comments

Posting Rules

Reply