Network perturbation analysis of gene transcriptional profiles reveals protein targets and mechanism of action of drugs and influenza A viral infection

Noh, Heeju; Shoemaker, Jason E; Gunawan, Rudiyanto

ProTINA requires a cell type-or tissue-specific PGRN as an input, which may hinder its application to analyze data from lesser studied organisms. In the case studies, we leveraged on the extensive online databases of proteinprotein interactions and TF-gene networks to manually curate PGRNs for human and mouse cells (33, 34, 36) . Alternatively, provided that a large dataset of gene expression profiles are available for the cell of interest, the PGRN could be inferred using existing network inference methods (73, 74) . Another potential limitation in applying ProTINA is the requirement for differential gene expression data for inferring the edge weights of the PGRN. While the minimum number for implementing ridge regression with a 3-fold cross validation (lowest fold in GLMNET) is three, the accuracy of the weights and thus the target predictions from ProTINA would generally deteriorate with lower sample sizes. Nevertheless, ProTINA was still able to provide reasonably accurate predictions using a total of 18 samples in the influenza A virus case study above.