Bayesian network for the evaluation of T/E data
A Bayesian network is a graphical probabilistic model used in forensic sciences because it provides a coherent structure to model the relationships among a set of causes and a set of scientific evidence while making possible to calculate the effect of knowing the truth of one proposition on the plausability of other propositions.
The Bayesian network shown here has been developed for the evaluation of evidence of steroid doping with T/EpiT data stored in an Athlete Steroidological Passport (ASP). Each circle represents a continuous variable, each rectangle a discrete variable, each arrow a causal relationship. These causal relationships are typically represented by models developed empirically from longitudinal clinical trials or from scientifically valid data found in the literature.
It is known that the between-subject variations of endogenous steroid concentrations measured in urine are strongly associated to the genetics of androgen disposition, and more particularly to a UDP-glucoronosyl transferase 2B17 polymorphism. This particular polymorphism has been well studied in endocrinology and maps of that genotype exist in function of the origin (ethnic or geographic) of the subject. For example, Japanese people are known to carry the deletion in both alleles (del/del) of this gene with no or negligible concentrations of testosterone in urine. This knowledge is formally taken into account in the Bayes network through the nodes “UGT2B17 genotype” and “ethnicity”.
An example of the evaluation of T/EpiT data can be found here.