From Association Analysis to Integrated Causal Inference
Momiao Xiong · 2018
This chapter introduces the assumptions for learning causal-effect models, and additive noise models for causal discovery of both qualitative and quantitative traits. It addresses the issues for integrating heterogeneous genomic, epigenomic, environmental, imaging, and phenotypic data into multilayer networks underlying disease and health. The approach to genomic analysis lacks breadth and depth and its paradigm of analysis is association and correlation analysis. Using association analysis as a major analytic platform for genetic studies of complex diseases is a key issue that hampers the theoretic development of genomic science and its application in practice. Although integrated genomic, epigenomic, and imaging data analysis take a huge number of variables into consideration in causal inference, there are still unmeasured causally relevant variables. The chapter discusses the additive noise model for distinguishing cause from effect. It examines the information geometric approach for inferring causal direction that exploits the asymmetry between cause and effect.