|
Abstract:
|
I will present a novel Bayesian causal discovery method for reverse engineering gene regulatory networks from purely observational genomic data. The discovered causal gene regulatory relationships are useful for predicting how a signaling pathway reacts when it is subject to certain perturbations (e.g., gene knock-in/knock-out) external to the system, which will be ultimately useful in designing targeted therapies in various genetic diseases. While the gold standard for causal discovery remains the controlled experimentation, large-scale gene knockout experiments are often too expensive to run and not widely available in public databases. By contrast, large-scale observational gene expression data are abundant in the public domain. Therefore, inferring unknown gene regulatory networks from purely observational data is desirable. Existing model-based causal discovery methods often make distributional assumptions such as IID Gaussian. The proposed method allows for inference of causality that is robust to the violation of these distributional assumptions, which is quite common in genomic studies. The utility of the proposed method is supported by theory, simulations, and real data.
|
