On public and private aspects of the adaptive immune system as revealed by statistical inference

The adaptive immune system can recognize unanticipated threats by maintaining a large diversity of T cells with different membrane receptors. The diversity is dynamic, with T cells constantly being produced with randomly-generated receptors. The randomness means that different individuals hold different repertoires of T cells, despite the fact they are able to defend against the same pathogens. An interesting aspect of this phenomenon is that certain receptors are “public”, appearing in almost any individual, while others are “private”, being unique to one individual.
We have learned how to infer probabilistic models that capture the statistics of the T cell ensemble carried by individuals. A key feature of these models is that individual immune cells are produced with different likelihoods, with the distribution of generation probabilities spanning many decades. The statistical power of our models enables us to compare samples from different individuals or compartments. We use the models to predict how many receptors (both in number and in specific receptor sequences) will be shared between any number of individuals, simply as a result of the biased generation statistics. We show results for mice and men and draw conclusions about the predictability of a stochastic immune system.