Immune System Modeling with Infer.NET

Graphical models allow scientific prior knowledge to be incorporated into the statistical analysis of data, whilst also providing a vivid way to represent and communicate this knowledge. In this paper we develop a graphical model of the immune system as a means of analyzing immunological data from the Manchester asthma and allergy study (MAAS). The analysis is achieved using the Infer.NET tool which allows Bayesian inference to be applied automatically to a specified graphical model.Our immune system model consists firstly of a hidden Markov model representing how allergen-specific skin prick tests (SPTs) and serum-specific IgE tests (SITs) change over time. By introducing a latent multinomial variable, we also cluster the children in an unsupervised manner into different sensitization classes. For 2 sensitization classes, the children who are vulnerable to allergies and have a high probability of having asthma (22%) are identified. For 5 sensitization classes, children in the first cluster, those who are vulnerable to allergies, have an even higher probability of having asthma (42%). The second part of the model involves using the inferred sensitization class as a label and 8 exposure variables in a Bayes point machine. Using multiple permutation tests, we conclude that the level of endotoxins and gender have a significant effect on a child's vulnerability to allergies.

[1]  Tom Minka,et al.  Expectation Propagation for approximate Bayesian inference , 2001, UAI.

[2]  Colin Campbell,et al.  Bayes Point Machines , 2001, J. Mach. Learn. Res..

[3]  A. Woodcock,et al.  The National Asthma Campaign Manchester Asthma and Allergy Study , 2002, Pediatric allergy and immunology : official publication of the European Society of Pediatric Allergy and Immunology.

[4]  Charles M. Bishop,et al.  Variational Message Passing , 2005, J. Mach. Learn. Res..