Finding cell type specificity

[waltno]

Hi

Thank you so much for releasing this awesome package! I've been able to use all functions/analyses with ease. Now, I am wondering if there is a good way to uncover/test which TF repressors/activators are specific to cells within a dataset, ie seurat clusters, experimental groups, cell type, etc. I've been able to make dual heatmaps of TF expression and chromvar scores--across clusters, but it'd be nice to have an integrated visualization and approach.

I am also wondering if running FigR on subsets of cells could work (by cluster or exp group for ex). I know this analysis is more powerful experiment wide..

Any insight would be helpful! Thank you!

[vkartha]

Hi there! Thank you for your interest in using our package. Regarding your question of specificity - this is indeed an area of our active work, and hopefully will be incorporated into future implementations of the FigR framework/package (i.e. focusing on metrics other than correlations). Given our approach is covariance based, we are limited to finding key activators and repressors across single cells that are input to the functions (both peak-gene testing and TF-gene testing). One thing you can always do is separately come up with a list of DORCs/ TFs that are indeed cluster (cell type / condition) specific (based on differential expression or DORC accessibility), and then intersect that with the final list of TF-gene associations to then see a filtered, context specific network (this is not the same as re-running FigR only on a subset of cells, which may give you different results given if the cells are very similar, the covariance across them will be low, and less likely to detect variable features that are significant across cells). Will add more information once we get further along with integration methods that will let you 1) score cells for regulation modules determined using FigR GRN 2) look into specificity of TF-gene relationships for different clusters.

Thanks again for your comment!