Fig. 1.

Overview of the SPONGE workflow. (A) Predicted and/or experimentally validated gene–miRNA interactions are subjected to regularized regression on gene and miRNA expression data. Interactions with negative coefficients are retained since they indicate miRNA induced inhibition of gene expression. (B) We compute sensitivity correlation coefficients for gene pairs based on shared miRNAs identified in (A). (C) Given the sample number, we compute empirical null models for various gene–gene correlation coefficients (k) and number of miRNAs (m). Sensitivity correlations coefficients are assigned to the best matching null model and a P-value is inferred. (D) After multiple testing correction, significant ceRNA interactions can be used to construct a genome-wide, disease or dataset-specific ceRNA interaction network