Computational Modeling of Immune Signals

• Main Author: Starzl, Ravi
• Format: Others
• Published: Research Showcase @ CMU 2012
• Subjects: computational immunology; immune signal modeling; transplant tolerance; toleragenic therapy; cytokine network; immune monitoring
• Online Access: http://repository.cmu.edu/dissertations/339; http://repository.cmu.edu/cgi/viewcontent.cgi?article=1339&context=dissertations

The primary obstacle to enabling wide spread adoption of composite tissue transplantation, as well as to improving long term solid organ transplant outcomes, is establishing a personalized medication regimen optimizing the balance between immunosuppression and immune function the individual minimum effective level of immunosuppression. Presently, the clinical gold standard for monitoring immune function is histologic inspection of biopsy for tissue damage, or monitoring blood chemistry for signs of organ failure. These trailing indicators reflect damage that has already accumulated, and are of little use in proactively determining the immunologic state of a patient. Samples collected from small animal surgical models were used to quantify the amount of immune signaling protein present (cytokines and chemokines) under various experimental conditions. Patterns in protein expression that reliably discriminate amongst the groups were then investigated with statistical inference methods such as the logistic classifier, decision tree, and random forest, operating in both the original feature space and in transformed feature spaces. This work demonstrates computational methods are effective in elucidating and classifying cytokine profiles, allowing the detection of rejection in composite tissue allografts well in advance of the current clinical gold standard, and shows that the methods can be effective in solid organ contexts as well. This work further determines that cytokine patterns of inflammation associated with rejection are specific to the structure and composition of the tissue in which they occur, and can be distinguished from immune signaling patterns associated with unspecific inflammation, wound healing, or immunosuppressed tissue. Clinical translation of these findings may provide novel computational tools that enable physicians to design personalized immunosuppression strategies for patients. The methods described in this work also provide information that can be used to investigate the biological basis for the observed immune signaling patterns. Further development may provide a computational framework for identifying novel therapeutic strategies in other pathologies.