log transformation before imputation (impute.knn) in proteomic data?

Question

[Lila M]

Hi all,

I am starting work with proteomic data and I am quite new , the first challenge I am approaching is to deal with NAs values and run imputation. My data has been normalized but it is skewed eg:

Min.   1st Qu.    Median      Mean   3rd Qu.      Max.      NA's 
220      8123     18080    702594     48181 132186128       162 

So my first temptation would be to log-transform it to have a closer normal distribution and then run imputation using the log-transformed data.

Is this a valid approach or would be better to run imputation in the 'skewed' data and then log transform the results?

I also appreciate if someone could point me to some extra documentation that support this decision!

Thank you!

Answer 1

[KABILAN]

By which normalization method you have applied to the dataset? Since normalization and imputation methods are data-specific in nature, you have to choose appropriate normalization and imputation methods for the dataset. I suggest you try the shiny app lfproQC (http://omics.icar.gov.in/lfproQC) which will suggest the suitable method of normalization and imputation methods for your dataset.