I am currently performing a compositional analysis of scRNA data that includes samples from multiple datasets. For this, I used various tools, including scCODA, propeller, and scikit-bio. scCODA was difficult to set up, so we opted for propeller and scikit-bio. The results from both methods vary a lot. The p-values for the clusters differ between the two methods. Can anyone explain why this variation is happening, and which method will be a better choice? Thankyou!!