|Stochastic Systems Group|
A Graphical Model of the Immune System
In this talk, a graphical model representation of the immune system will be presented. The data was obtained from the Manchester Asthma and Allergies Study (MAAS) and it includes 1186 subjects. The model combines physiological measurements, exposure variables and genetics in the form of Single Nucleotide Polymorphisms (SNPs). Using a Hidden Markov Model (HMM), the available physiological measurements such as specific skin prick tests and specific IgE tests to a set of allergens allows us to infer a latent acquired sensitization state for each subject and each allergen. Furthermore, our model allows us infer one multinomial latent variable per child to cluster the children in an unsupervised manner into different sensitization classes. For 2 sensitization classes, this clustering clearly partitions the children into those who are vulnerable to allergens and have a higher probability of having asthma (23%) and those who are not vulnerable to allergens and have a lower probability of having asthma (3%). The second part of the model involves using the inferred sensitization class as a label and the exposure variables as covariates in a Bayes Point Machine (BPM) model. Lastly, we explore genetic-environment-sensitization interactions using the SNP data. This may reveal new potential SNPs and environmental covariates that work in tandem to increase the risk of asthma, which may increase the understanding of the asthma-related biological pathways.
Problems with this site should be emailed to firstname.lastname@example.org