[SCAVENGE] Rule of thumb for SCAVENGE analysis

Use a subset of variants instead of all variants of a trait

1. It is totally feasible to use a subset of variants of a trait of interest to evaluate if this subset has a different cell type enrichment or a different strength of enrichment compared to the whole set of variants, as SCAVENGE can overcome sparsity issues in the analysis. A couple of thoughts in terms of this subset analysis.

2. The variants with greater posterior probabilities (PP) (for example, >0.5) affect the enrichment a lot (although we generally used all the variants with PP>0.01 as input). So PP is not a good filtration for generating the variant subsets for an unbiased analysis. Although we can reduce the effects of sparsity, the number of variants in the subset should not be too small since if there is no overlap between the ATAC signal and variants, the initial g-chromVAR analysis will produce NA for most cells. The following network propagation is not allowed. The minimal variant number should be explored.

Interpretation and visualization of SCAVENGE analysis

1. Both the fold change of TRS and the percentage of enriched cells can be informative. For the fold change comparisons, using the TRS rather than network propagation scores make more sense, as network propagation scores were scaled into the 0-1 range in all situations.
   The median of TRS of a specific cell type is not the only option, as we usually see a large variant in TRS of a cell type, sometimes a number between 5%-20% TRS quantile in a cell type-specific manner could be better.

2. For comparison of the percentage of enriched cells from different traits, additional criteria can be added other than the empirical P value, which is the TRS that can be viewed as a variant of Z score. So cutoffs of both p-value (0.05) and TRS (1.5 or 3, or a reasonable value uniformly used across subsets) can be applied to select the enriched cells. This is not shown in the original paper as the comparisons of different traits were not involved.