Neural networks elucidate T cell priming conditions for adoptive transfer

Both host cytokines and microbial metabolites can influence the differentiation of naïve T cells near the gastrointestinal (GI) tract. While some differentiated T cells mitigate inflammation, others promote it; thus T cells near the GI tract exist in a delicate homeostasis that allows for simultaneously tolerating commensal microbes and destroying harmful pathogens. Environmental stimuli coupled with genetic predisposition can disrupt this homeostasis and inflammatory bowel disease (IBD) ensues. A promising treatment for IBD involves culturing a patient´s naïve T cells, differentiating them into anti-inflammatory T cells, and transplanting them back into the original patient in a process known as adoptive transfer. One bottleneck in translating this technique from the laboratory to the clinic is in the determination of appropriate stimulation conditions prior to transplantation. Since the biochemical pathway underlying this differentiation is largely unknown, only data driven models can be used to model the effect of various stimulation conditions on the percentages of anti-inflammatory Tregs and pro-inflammatory Th17 cells. This work develops such models where a special emphasis is placed on determining empirical models that maximize the prediction accuracy. Neural network models are used due to both their flexibility for choosing model structures and their ability to reflect complex phenomena.