search

HelenaLC / muscat

Multi-sample multi-group scRNA-seq analysis tools
166 stars 33 forks source link

The p_adj.loc results #72

Open yuyingxie opened 3 years ago

yuyingxie commented 3 years ago

Hi muscat team,

I ran the mmDS function with the method of 'dream2' and got the following results. Screenshot 2021-06-24 141140 The p_adj.loc results are so close to 1, which seems incorrect. Do you have any suggestions to fix it?

Thanks

HelenaLC commented 3 years ago

I don't quite follow. What makes you think the adjusted p-values are incorrect? One cannot tell from looking at 20 results, but I see only insignificant p-values & lowish effect-sizes (logFC), so the results look reasonable.

lindsaynhayes commented 3 years ago

I have a conceptual follow-up to this question. I have run into a similar challenge. I have DEG with significant p_adj.loc values but the logFC was low (ie < abs(0.5)). However there were other genes in which the logFC was high (ie > abs(1)) but were insignificant. When I plot these they do look convincingly different. I have been confused by why the genes with the most obvious looking fold change when plotted are insig and the ones that are sig are less convincing when plotting the expression. In bulk analysis, when this happens I can usually identify the 1 or 2 samples that have a very high/low expression that is driving the large fold change. However in the single cell data this kind of outlier driving the fold change is not as clear. If there are a lot more zero cells in one group would that cause a high fold change but insignificant statistical test? Is there a way to get around that? A big change in zero cells is biologically meaningful.