q2-gemelli - auto components

[arwqiime (ARW)]

Hello @cmartino
I have previously used q2-deicode to generate Aitchion-based distance matrices and plotted them with q2-emperor. Since I had issues installing deicode to q2-amplicon-2024.10, I switched to gemelli (assuming that this is the successor of deicode according to the github information).
I observed with a new dataset some kind of clustering with gemelli as described in this previous post:

Peculiar clustering of samples in DECODE RPCA biplot

Hi @cmartino! After playing around with the --p-min-feature-frequency and --p-n-components, I found that increasing the rank to 6 or using auto-rpca actually helps a bit but setting the minimal frequency to 10 didn't help. rpca-biplot-autorpca.qzv (1.2 MB) rpca-biplot-minFreq10.qzv (1.2 MB) rpca-biplot-ncomp6.qzv (1.2 MB) Do you think the plots with an increased number of ranks look normal?

Since I gemelli does not contain a auto-rpca method, I manually set --p-n-components to 10 and obtained plot more similar to other plots (samples not lined up on a "line").
Is there a recommendation to find the smalles number of components, or should it be tested with manually definded components?

Best,

[cmartino (cameron martino)]

Hi @arwqiime,

Thanks for using RPCA! Yes, please use Gemelli, the deicode package is no longer supported, so all the bug fixes are in Gemelli. The auto-rpca feature was removed because it was not functioning properly. The default of --p-n-components=3 is sufficient for most users since most only plot/view the first three PCs anyway. Your --p-n-components=10 is also fine, but I would not go much higher than that since it will increase the runtime and violate the low-rank assumptions (which all dim. reduction methods make).