In this chapter, graphical models are introduced and used as a natural way to formulate and address problems in genetics and related areas. Local computational algorithms on graphical models are presented and their relationship with the traditional peeling algorithms discussed. The potential of graphical model representations is explored and illustrated using examples in linkage and association analysis, pedigree uncertainty, forensic identification, and causal inference from observational data.
INTRODUCTIONGraphs appear in a number of different contexts in genetics to convey information, e.g. about population development and evolution, and relationships between genes and individuals. This chapter focuses on the role of probabilistic graphical models (Lauritzen, 1996) within genetics, and in particular, on aspects of genetics which involve pedigree analysis, i.e. the analysis of genetic information among related individuals.Probabilistic graphical models have their origin in genetics, in path analysis (Wright, 1921;1923;1934), which explicitly studies the propagation of hereditary properties through a family tree. They form a natural general framework to express and manipulate a number of important aspects of statistical genetics, e.g. computational algorithms such as 'peeling' (Elston and Stewart, 1971;Cannings et al., 1978;Lander and Green, 1987), but have applications beyond that; e.g. in forensic genetics where complex issues of identification can be naturally expressed in terms of graphical models, in genetic epidemiology where the notion of Mendelian instruments helps to identify causal effects of genes, and in the study of regulatory networks, where graphs are naturally suited 808 Handbook of Statistical Genetics, Third Edition . E dited by D . J. Balding, M . Bishop and C. Cannings.