A0755
Title: Inferring associations along the causal chains in a network
Authors: Chen-Hsiang Yeang - Academia Sinica (Taiwan) [presenting]
Abstract: In a large system where the relations between entities are represented as a network, the effects of perturbing entities are propagated along the paths in the network. Many perturbations and responses may occur concurrently in the same experiments or observational events. Therefore, it is generally difficult to distinguish the causal relations from the spurious associations in such a system. Moreover, although the causal effects can, in principle, propagate along all directions permitted by the network structure, in reality, the majority of the causal effects are likely mediated by the paths restricted to a compact subnetwork. Finding this core subnetwork from the massive amount of association data also remains an open problem. Several algorithms are proposed to tackle these two open problems. First, a model selection procedure is built, which prioritizes the putative associations by their path lengths connecting the perturbations and effects and iteratively incorporates the candidate associations that best fit the data. Second, a network diffusion model is constructed, which assigns edge weights according to the paths traversing perturbations-effects pairs, and develops an algorithm to extract the core subnetwork with high edge weights. These two algorithms are applied to the multi-omics data of 33 cancer types in The Cancer Genome Atlas (TCGA) and identify the Integrated Hierarchical Association Structure (IHAS) within and across the cancer types.
