search

morris-lab / CellOracle

This is the alpha version of the CellOracle package
Other
306 stars 54 forks source link

Using mouse ATAC Atlas with mm10 #48

Closed f6v closed 3 years ago

f6v commented 3 years ago

I'd like to compare the results I get from CellOracle with my ATAC data and with mouse ATAC atlas. Since my data is mapped to mm10 I'd like to run liftover on the atlas data. However, is it correct to take the output from co.data.load_TFinfo_df_mm9_mouse_atac_atlas() and translate the genomic coordinates, or should I start from all the way from the normalised counts?

EDIT: I think the coordinates of TFinfo_df aren't taken into account when oracle.import_TF_data is used, are they? Does this idea of translating mm9 to mm10 make sense to you at all?

Thanks!

KenjiKamimoto-ac commented 3 years ago

Hi f6v,

Thank you for your question! The answer depends on the aim of your analysis.

(1) If you aim to compare GRN analysis results, you do not need to translate the data from mm9 to mm10.

The genome coordinate information was required for the motif scan, but it is not used in subsequent GRN analysis; The GRN analysis just requires paired lists of TF and its target genes.

As you pointed out, the cordinates of TFinfo_df are not taken into account in most GRN analysis. Although the gene location information is still stored in the oracle object, You can just ignore this information when you do a GRN analysis. The peak cordinates of TFinfo_df is used for a different analysis.

(2) If you aim to compare TF binding genomic locations, you need to translate the data or you need to start ATAC-seq analysis from scratch.

In this case, I recommend to do the analysis from scratch. For example, please get fastq files of mouse ATAC Atlas data and align NGS reads into the same reference genome as you used for your ATAC-seq data. I think the TFinfo data need to be prepared in the same manner. I guess it is better to process the atac-seq data in the same pipeline if you want to compare ATAC-seq peak location or TF binding patterns.

f6v commented 3 years ago

@KenjiKamimoto-wustl122 thanks! I was meaning to compare GRN results, but both suggestions are insightful!