This presentation focus on feature selection between two (or more) conditions, in  XIC metaproteomics data. This type of data obtained by LC-MS/MS technologies display a large proportion of missing values, and the usual approach consists in  first imputing missing value, then applying a differential analysis procedure by treating equally imputed and observed values. We propose an alternative which circumvents missing data imputation via a combined test  that targets two types of behaviour:  either a difference in terms of missingness between the two conditions, or a difference of intensity among the sample in which the feature (protein, peptide) was actually observed.
I will propose an overview and a comparison of data imputation methods, notably using a data set including technical replicates, and compare these performances with our combined test procedure.