Hi! I’ve been using metaboanalyst statistical analysis for analyzing a large untargeted metabolomics dataset. As usual, there are 0 values (which are not necessarily zero, just below the detection limit). I am confused if I should be replacing these with NA now? In past versions, I’ve tested both and it yielded the same results (because I believe at the time it was just replacing both with 1/5 of the minimum value). Because of that, I stopped bothering to replace the zeros with NA, but it has come to my attention that MetaboAnalyst now offers different options for dealing with NAs vs zeros. For zeros, it seems to continue to just replace them with 1/5 of the minimum value, but if you select this option when plugging in NAs (which, in theory, should give you the same results), it gives you different results. I’m unsure which one I should be using, and any advice would be much appreciated!
This is a very good question!
Untargeted metabolomics data contains very high proportion of missing values. Our simple approach (developed for targeted metabolomics) – replacing them (0 or NA) with 1/5 of minimal positive values for each feature followed by normalization/transformation – could have large impact on the data (which is not desirable as they are not actual measured values). To address the issue, we added more advanced methods (scuh as QRILC) that have been developed to deal with such issue in microarray/proteomics. In such method, 0 and NA are treated differently.
In summary, use NA for missing values. Zero should not be used for missing values.
This topic was automatically closed after 8 days. New replies are no longer allowed.