Abnormal kBET Scores

[VishD17]

Hello,

Thank you for a great tool! We had some questions on interpreting the results. In the image below, the figure on the top represents integrated data, where species are well-mixed in. Yet, the rejection rate is 1 for all samples, suggesting that the local and global distribution of species labels are completely different.

On the unintegrated data, however, the rejection rate averages to 0.91. How should we interpret these results? We were under the impression that kbet observed should be high for unintegrated and low for integrated data. Thank you for the help!

[mbuttner]

HI @VishD17

thank you for sharing your results here. The integrated kBET result looks counter-intuitive to me, too. What is the neighborhood size that you use for your data? The other possibility is that the global distribution of your batches differs in each of the cell types in your data, which drives the kBET scores up (I assume that each of the "blobs" of integrated data is a cell type). You could repeat the kBET test per cell type and it should yield lower scores.
I hope that helps!

[VishD17]

Thank you so much for the timely reply @mbuttner! We will take a look again considering your suggestions.