Pathway enrichment analysis models (PEM) are biological inference approaches that leverage annotated bio-molecular functions for interpreting the underlying processes of gene expression profiles from high-throughput genomic data. Common PEMs neglect the interactions among the gene/proteins and regard the known annotated functions as simple lists, even though the interactions are essential components of biological systems. Disregarding the interactions in the standard PEMs potentially results in inaccurate inference, especially when focusing on the biological pathways, which are important sub-classes of the biological knowledge. Network-based PEMs are emerging methods that account for the interactions in the biological networks to produce more informative functional interpretations. However, the methodologies that are used in the current network-based PEM do not necessarily capture the key features of the topological organization of pathways, including the upstream/downstream characteristics.This research study devises a pathway enrichment analysis by using a novel graph model, Source/Sink centrality (SSC), to capture the network organizations in pathways effectively. The key idea of SSC is to measure the importance of a gene in both upstream and downstream of a pathway while accounting for the temporal/biochemical order of the interactions. We use SSC to derive a topological statistic for the importance of a given set of genes in the network of a pathway, and use this topological statistic to construct a network-based PEM, called Causal Disturbance Analysis (CADIA). The performance of CADIA is validated by showing that it uniquely produces relevant critical interpretations in multiple sets of experimental data, while other PEMs fail to do so. We also use synthetically generated data to evaluate the specificity of CADIA in detecting pathway enrichments.This research study also shows an exploratory evaluation of the SSC by hypothesizing that it can capture the topological organization of \textit{a priori} known important genes. To this end, we investigate a battery of standard graph centrality models and their novel SSC extensions for describing the organization of cancer genes in the human pathways. From multiple perspectives, we show that the SSC extensions can distinguish between the topological positions of cancer and non-cancer genes. These results show that the SSC methodology contribute to the biological inference methods, as it can effectively capture the topological organization of a particular class of important genes in the biological pathways.