Comparison of strategies for missing data. Effect of the treatment of missing LC-MS peaks on protein significance analysis in the study of breast cancer cell lines, for the subset of 140 proteins where at least one feature is missing entirely in at least one condition. (a) Results of comparing protein abundances between cell line types after six hours of normoxia. (b) Results of comparing protein abundances between cell line types under normoxia, on average over the two exposure times. Each circle shows the number of differentially abundant proteins detected with three treatments of features with peak intensities missing in an entire condition: (1) imputation of the background intensity, (2) assuming no feature interferences, and (3) removing the feature from the dataset. The two comparisons have the same biological interpretation, but (b) doubles the number of biological replicates.
Clough et al. BMC Bioinformatics 2012 13(Suppl 16):S6 doi:10.1186/1471-2105-13-S16-S6