should we predict peak-gene pairs across all cells? Not restricted to a specific cell type?

[YichaoOU]

Hello,

Thank you for the great tool.

One question I have after reading the paper, also looking the code, is seems that the peak-gene pairs are predicted using cells from one cell type (cluster). Should we do such prediction using all cells? In this way, the association may be more significant? because of more cells?

Thanks, Yichao

[saorisakaue]

Hi Yichao, We are recommending to create links within each of the broad cell types (e.g. T cells). We intensively benchmarked this in the paper and described it in the Supplementary Note 3. I cite some summary of the main conclusions below:

While association across cells from different cell types might increase the number of significant peak–gene linkages due to greater variance in chromatin accessibility and gene expression, this strategy could yield false-positive enhancer–gene associations by increasing the chances of connecting enhancers that are merely ‘correlated’ with gene expression (Extended Data Fig. 10a). By simulation and real data analyses, we confirmed that the cell-type-specific analysis was better calibrated to reject false enhancer–gene links by correlation and powered to detect experimentally validated enhancer–gene links, while it was less powered to detect promoter–gene links when compared with multiple-cell-type analysis (Supplementary Note 3 and Extended Data Fig. 10b–e).

I hope this helps! Saori