A noise model allows quantitative interpretation of immune responses at repertoire scale

Longitudinal repertoire sequencing allows for the measurement of the dynamics of adaptive immune responses at scale. However, disentangling the true dynamics from sampling noise remains a major challenge. Here, we have built a mechanistic statistical model, which we validated against replicate data, that captures the multiple steps in the measurement process. Our model identifies the broad distribution of the number of mRNAs contributed by single cells as a parsimonious explanation for the observed overdispersed read counts. This mechanistic insight naturally leads to efficient computational methods for identifying significantly expanding lymphocyte clones. We apply these methods to investigate T cell clonal dynamics following Yellow Fever vaccination and acute SARS-CoV-2 infection, and contrast our findings with predictions of simple dynamical models.