Abstract

A patient’s tumour microenvironment, and especially the interaction between a patient’s cancer and their immune system, plays a crucial role in disease progression, response to treatment and prognosis. This interaction should therefore be considered to meaningfully capture a person’s disease through the creation of digital twins.   T cell receptor (TCR) sequencing data, while providing an invaluable window into cancer-immune system co-evolution from the perspective of the immune system, are notoriously difficult to model and analyse. This is primarily due to the vast diversity in sample sizes between different TCR samples, the massive down-sampling step from population to sample, and the unique fingerprint a person’s TCR repertoire possesses. However, addressing these issues is key to being able to create a ‘digital twin’ of a person’s immune response to cancer, and central to understanding their disease.   To this end, we propose a mathematical/statistical framework that is able to identify whether related TCR samples, such as those from different regions of a patient’s tumour, or different timepoints, are likely to have come from the same repertoire (population), or whether we can detect significant changes between them (such as expanding T cell clones), for example, in response to treatment or metastases.   We have designed a mixed effects model, where TCR clone size frequencies are modelled as a random effect with different dispersion parameters for each sample. The model tests whether there is evidence for differences in clone frequencies between samples, beyond those due to sample size. We show that our model can identify inherently similar / different repertoires generated from simulated data and from an experimental dataset of technical repeats, then demonstrate its use on longitudinal and spatial metastatic colorectal cancer patient data. Our model mitigates the need for steep subsampling and the associated data loss, which is currently the most commonly used method for immune repertoire sample comparison.   As next steps, we will build on the model to further interrogate the cell dynamics that could lead to the observed clone size distributions. These could include, for example, the growth rate of clones and measurement of the influx of new clonotypes. This will ultimately result in a mechanistic model of changes in the immune repertoire through these processes.