Hi Ruochi,
Thanks for this great toolset!
The task I am trying to do with Higashi/Fast-higashi is to generate imputed contact matrices in high resolution (ideally 5k or 10k) in the pseudo-bulk level. I am wondering if this pipeline can be used for this goal:
- Use Fast-Higashi at low resolution (100k or even 500k) to generate embeddings and cell type annotations efficiently.
- Feed the pre-trained embeddings from Fast-Higashi into Higash model but with high resolution (5k or 10k).
- Extract imputation results from Higashi output.
- Use Merge2Cool tool to merge single cell high resolution imputed contact maps for a given cell type.
If this pipeline can work, I am also wondering if it saves resources at all or this is not different from using Higashi only for all steps.
We use snm3c for all our projects so that we usually create cell type annotations using the methylation data. Do you recommend running imputation/TAD/compartment calling at single cell level or directly using tools for bulk data on the pseudobulk merged contact files? I'd like to hear your thoughts and would appreciate any help you can provide.
Best,
Bing
Hi Ruochi,
Thanks for this great toolset!
The task I am trying to do with Higashi/Fast-higashi is to generate imputed contact matrices in high resolution (ideally 5k or 10k) in the pseudo-bulk level. I am wondering if this pipeline can be used for this goal:
If this pipeline can work, I am also wondering if it saves resources at all or this is not different from using Higashi only for all steps.
We use snm3c for all our projects so that we usually create cell type annotations using the methylation data. Do you recommend running imputation/TAD/compartment calling at single cell level or directly using tools for bulk data on the pseudobulk merged contact files? I'd like to hear your thoughts and would appreciate any help you can provide.
Best,
Bing