ATAC components

[SarahOuologuem]

Hi,

missing ATAC components include:

• ATAC-specific QC metrics (TSS, nucleosome_signal,...) see https://www.sc-best-practices.org/chromatin_accessibility/quality_control.html#calculating-qc-metrics and https://stuartlab.org/signac/articles/pbmc_vignette.html#computing-qc-metrics

• scATAC doublet detection tool, e.g. AMULET or scDblFinder. see: https://www.sc-best-practices.org/chromatin_accessibility/quality_control.html#doublet-detection

• Binarize peak count matrix

• TF-IDF normalization

• HVF selection

• LSI as dimensionality reduction

[VladimirShitov]

Thank you, it makes sense! What about GRN inference? From this page: https://www.sc-best-practices.org/chromatin_accessibility/gene_regulatory_networks_atac.html

[SarahOuologuem]

I think it's nice! But as far as I see, that's for the multimodal case of RNA+ATAC. I would say, if we want to start with the unimodal case, then that's for later.
I can start with the LSI component

[bio-la]

thanks @SarahOuologuem for working on LSI!
please check @DriesSchaumont's last update on the "setup" component that includes PCA, umap and Neighbours, #396
you would need to add LSI as an additional dimred and test that when the component is called with one of the 2 dimreds or both, the outputs are consistent.

[VladimirShitov]

I'll take QC metrics and doublet detection :)

[VladimirShitov]

We also need:

• HVG selection (literally scanpy function, so we can reuse existing components)
• Normalization and log-transform (scanpy functions)
• Neighbors graph building (scanpy function)
• Clustering (scanpy function)
• UMAP or other dim red methods (scanpy function)
• Marker peaks search – this we should implement from muon

[VladimirShitov]

Double check HVG selection: it might use a slightly different approach due to high sparsity and the number of features