Figure 1

The overall pipeline of NBIA. The input consists of m independent datasets and n genes. Step (1): calculate effect size (Hedge’s g) for each gene in each study. Step (2): combine effect sizes for each gene across multiple studies using the REstricted Maximum Likelihood (REML) algorithm. Step (3): compute the z-score (\({z}_{i}=\frac{{\mu }_{i}}{{\sigma }_{i}}\)) and calculate the left- and right-tailed p-values (ep_il and ep_ir) using the standard normal distribution. This ends the first module. Step (4): perform hypothesis testing at gene level using empirical Bayesian statistics. For gene i^th and dataset j^th, the left- and right-tailed p-values obtained from the Bayesian test are bp_ijl and bp_ijr. Step (5): combine the one-tailed p-values for each gene, i.e., bp_il = addCLT(bp_i1l, …, bp_iml) and bp_ir = addCLT(bp_i1r, …, bp_imr). This ends the second module. Step (6): combine Bayesian p-values with the p-values of the effect size using maxP, i.e. p_il = max(ep_il, bp_il) and p_ir = max(ep_ir, bp_ir). Step (7): choose genes that are significantly impacted from both hypothesis testing and effect size perspectives using FDR-adjusted p-values (1% threshold by default). This ends the third module. Step (8): compute the perturbation factors for NBIA-prioritized genes and pathways. Step (9): identify impacted pathways using impact analysis.