search

biocore / DEICODE

Robust Aitchison PCA from sparse count data
Other
33 stars 17 forks source link

Still having issue with arrow bias in biplot #60

Open manasson opened 4 years ago

manasson commented 4 years ago

Hello,

I previously left a comment about having issues with arrow bias in the deicode biplot (scher-lab #56). The recommendation was to re-install with the dev version, which I did. However, re-installation did not fix the issue.

The plot after re-installation looks like this: image

The code looks like the following:

run robust PCA

qiime deicode rpca \ --i-table $SKINDIR/outputs/06_tables-filtered_f/table_samples_silva-132-99-full-download_feature-filt.qza \ --p-min-sample-count 1000 \ --p-min-feature-count 10 \ --o-biplot $SKINDIR/jobs/02_deicode_f/ordination.qza \ --o-distance-matrix $SKINDIR/jobs/02_deicode_f/dist-matrix.qza \ --verbose

create biplot

qiime emperor biplot \ --i-biplot $SKINDIR/jobs/02_deicode_f/ordination.qza \ --m-sample-metadata-file $SKINDIR/inputs/01_map/map_skin.pso.psa_samples_for.analysis_20200629.txt \ --m-feature-metadata-file $SKINDIR/outputs/05_taxonomy_f/taxonomy_silva-132-99-full-download.qza \ --o-visualization $SKINDIR/jobs/02_deicode_f/biplot_20-features.qzv \ --p-ignore-missing-samples \ --p-number-of-features 20 \ --verbose

Thank you again!

cameronmartino commented 2 years ago

Hi @manasson,

I see this occur when there is a large change in the absence/presence of certain taxa. For example, when comparing those with or without antibiotics or with/without IBD the arrows tend to point all in one direction. This is because the main driving factor is the loss or bloom of species/features. I would suggest taking those features/arrows and looking at the absence/presence between samples, for example, tested with a Chi-squared test between groups.