Hi, I am using MetaboAnalyst for untargeted metabolomics data. I have a large dataset which inlcudes 19000 features or something. I have 0 values in my data. When I perform the data filtering step, I wanted to check the dataset from download section. When I look for the processed data, I observed there are some different values for my 0 values. Does this data filtering include any missing value estimation and imputation step or something like this? I am confused about this because I did not perform the missing value estimation step. What are those values which are replaced with 0? Thanks.
Data processing is to prepare for normalization and statistical analysis. Since zero values will cause trouble in log transformation, they will be replaced by the 1/5 of the smallest positive values of that feature (i.e. replacing zero with detection limit).
The approach should not lead to distributional changes unless your data contain a large percentage of zeros. If this is the case, you should do formal data filtering and missing value imputation.
Hi Prof. Xia,
I have a follow up question to this answer.
I have an MSDIAL output with test samples and blank samples. There are couple of blank samples have high intensity values for some metabolites. When I substrate blank’s value from samples I get negative values for these samples. Since these are biologically meaningless I replaced them with NA.
In the same data set there are some “0” values. Is it ok to replace them as NAs too?
How to manage 0 and NAs in downstream analysis.
Thank you in advance.