In integrative gene expression analysis when two expression matrices from different datasets are merged together, they are usually merged by common genes in both expression matrices while all other genes are excluded. Common methods of batch effect removal, like empirical Bayes method (
ComBat function from R package
sva), and of searching for differentially expressed genes, like generalized linear models from
limma R package, assume completeness of data for each gene and each sample. In case of integration of several datasets the result is often a trade-off between an increased statistical power of the combined dataset and the reduced amount of the observed genes. This problem impedes large scale dataset integration.
Is there a method to integrate expression data and search for differentially expressed genes without throwing away genes where expression data are absent in at least one of the datasets?