4/ ACCESS-ATAC significantly improves TF binding site prediction, and in single-cell ACCESS-ATAC experiments, we resolve TF binding profiles at equal sensitivity to ATAC-seq with 2-3x fewer cells.

I would trust that BPNet does better than linear models, but I am not sure about the LM models, which are often published by only compare among themselves. See the study on DL models in insilico pertubation: www.biorxiv.org/content/10.1...
Maybe a 1-layer BPnet would do the job here?

Very nice benchmarking Anshul! I just missed a weak baseline (linear model?). Experience shown that they might still be better or as good as a lot of DL/LM models.

SIGURD is implemented as an R package with associated pre-processing scripts. It is available on github.com/CostaLab/sig.... Amazing collaborative work with M. Kalmer, S. Koschmieder, @brummendorf within the #CRU344 #genotyping #singlecell #MPN 5/5

We build on MAEGATK for delineation of mitochondria derived clones. We observed an increase in clonality in MPN compared to HC, as supported by diversity measures. As before, we can functionally characterize clones with GSEA and enrichment analysis 4/5

We demonstrated SIGURD in scRNA-seq and JAK2V617F amplicon data from Myeloproliferative Neoplasms. Combined analysis with scRNA-seq allowed us to find an enrichment of JAK2V617F cells in late erythroid cells and an increase in heme metabolism 3/5

SIGURD combines distinct genotyping tools (cellSNP-lite, VarTrix, MAEGATK) into a single reproducible framework. Genotyping results are stored in a data container and can be used for integrative analysis with single cell RNA #Seurat 2/5

✋