I'm trying to recreate the CONICS analysis for this paper, in particular for the scATAC-seq samples. Like in the paper I'm using log2(CPM/10) for the Gene accessibility scores of the 7 samples that are available through the EGA (SF11979 seems to be only available as RNA). I'm using the chromosome arms from the tutorial as CNV list. Unfortunately, it seems like CONICS isn't picking up on the different tumor populations but mostly finds some very small populations that are marked as seperate (<1%) . I do see some clear translocations in the heatmap though.
I'm already filtering more stringently than in the paper (i.e. FRiP > 0.2 and min. passed filter > 5.000) and the cells don't seem to be outliers nor do they pop up as seperate populations in my tsne.
I was wondering if you could share your script that was used in the paper and some meta-data with the tumor/healthy classification. The paper only mentions setting the likelihood-ratio test <0.001 and a difference in Bayesian Criterion >300, perhaps there are more things that need to be taken into account when working with scATAC-seq that I am overlooking.
Hope to hear from you soon and I'm liking this approach a lot and hoping to apply to my own pipeline as well!
Hi @diazlab and @soerenmueller ,
I'm trying to recreate the CONICS analysis for this paper, in particular for the scATAC-seq samples. Like in the paper I'm using log2(CPM/10) for the Gene accessibility scores of the 7 samples that are available through the EGA (SF11979 seems to be only available as RNA). I'm using the chromosome arms from the tutorial as CNV list. Unfortunately, it seems like CONICS isn't picking up on the different tumor populations but mostly finds some very small populations that are marked as seperate (<1%) . I do see some clear translocations in the heatmap though.
I'm already filtering more stringently than in the paper (i.e. FRiP > 0.2 and min. passed filter > 5.000) and the cells don't seem to be outliers nor do they pop up as seperate populations in my tsne.
I was wondering if you could share your script that was used in the paper and some meta-data with the tumor/healthy classification. The paper only mentions setting the likelihood-ratio test <0.001 and a difference in Bayesian Criterion >300, perhaps there are more things that need to be taken into account when working with scATAC-seq that I am overlooking.
Hope to hear from you soon and I'm liking this approach a lot and hoping to apply to my own pipeline as well!