Yes, the way Cicero normalizes counts and computes pseudo cells from UMAP coordinates was really deleterious in our evaluation..

Maybe improvement will come with scATAC atlases, to compensate for data sparsity! 😊

Very happy to see this work finally online 🥳

It was really interesting to write a short manuscript mixing method presentation & broader insights on data preprocessing. 📝

Thank you to my co-supervisors @cantinilab.bsky.social & @juliosaezrod.bsky.social for their feedback on this side work! 🫶

5/5

CIRCE is a Python package based on #AnnData and fully compatible with the #scverse environment.

You can extract DNA region modules & visualize CIRCE's interactions over gene locations.

CIRCE also facilitates #CellOracle or #HuMMuS usage, making it unnecessary to run Cicero (code in R) first.

4/5

We explore the impact of preprocessing on DNA interactions inference. 👩‍🔬

We evaluated scATAC-seq preprocessings using promoter capture Hi-C data. 🧬

TLDR: Best performance came from single-cell inputs directly and from CIRCE metacells! 📊
In contrast, count normalization had a negative impact.

3/5

We compared CIRCE & Cicero on 2 datasets, demonstrating near identical results from the same input.

Both use pseudo-cells to reduce sparsity. CIRCE proposes a new strategy, whose output is closer to the single-cell profiles. 🛠️

On average, CIRCE uses 5x less memory and runs 150x faster ! 📈

2/5

Congrats Pau for all this work !! 🎉🎉

I learned a lot from being around you (and I still do) 😇
And I'm really looking forward to seeing what you'll bring to the field in the coming years 🔮

I would love to be in too ! 😊

Hi, would like to join the science feed too please ! :)

I'm an MD-PhD student currently working on single cell method development to predict molecular regulations.
orcid.org/0000-0001-87...