search

dviraran / xCell

Cell types enrichment analysis
184 stars 62 forks source link

xCellSignifcanceBetaDist intrepretation #67

Open Shubhra1994 opened 2 years ago

Shubhra1994 commented 2 years ago

Hi, Thanks for developing this. So I have been trying to intrepret the results if xCellSignifcanceBetaDist and the output contains many cell types where the pvalue defaults to 0. I have a large dataset of approximately 800 patients.

When I do not use this test and select the cell types based on biological knowledge my Xcell scores are very low around 0.3 at maximum.

Could you let me know why this behaviour maybe coming up in terms of defaulting to 0 or low Xcell score.

Thanks!

dviraran commented 2 years ago

Hi,

xCellSignifcanceBetaDist is an attempt to provide some confidence as to whether the scores are significantly different from 0. It is based on the fact that some of the cell types signatures have higher error rates than others.

It doesn't matter how big your dataset is, it just uses predefined estimations of the error rate to calculate the p-value. I am not completely sure I understand the questions themselves. P-value of zero is good, it means the scores are higher than the error rate.

I hope this helps.

Best, Dvir

Shubhra1994 commented 2 years ago

Thanks Dvir,

Make sense, but why do you reckon that the xCell scores are low (0.3 or so) when using the cells which have known biological function in the disease, and I am filtering only these cells for estimation.

Do you think the score will improve if I take cells which are predicted to have scores higher than error rates as calculated according to xCellSignifcanceBetaDist.

dviraran commented 2 years ago

0.3 doesn't seem so low to me, its actually pretty high.

I don't think it will have much of an effect, but you can try. You should get better results if you run it only for cell types that are expected to be in the mix.