The results of STRUGGLE, show that theoretical approaches can extract more from data than standard methods, changing how we think about public and private responses. They also resurrected an interest in thinking about the immune system as a sensory, evolving non-equilibrium system. Immune repertoires hold information about past infection history. The results of this project show that we are able to extract this information.
We proposed different methodologies for identifying responding immune cells from snapshots, longitudinal and cohort data. We developed software packages and algorithms (ALICE, OLGA, SONIA, SONNIA, SOS) and characterised population level differences. We also developed efficient algorithms that identify shared clonotypes between individuals in human T-cells and B-cells. Our methods proved very timely for the pandemic.
We developed machine learning (ML) methods for epitope presentation (that can be used to personalised datasets with very rare HLA alleles) and predicting the binding between TCRs and their cognate epitopes (TULIP). Our work on antigen-antibody models reliably identifies epistasis, showing that epistatically interacting sites contribute substantially to binding. In addition to negative epistasis, we report a large amount of beneficial epistasis, enlarging the space of high-affinity antibodies as well as their mutational accessibility.
In more theoretical work, we explored this angle and developed new inference methodology for collective dynamics. We developed theoretical models for viral evolution in a population of host immune systems and used information theoretic approaches to quantify prediction for viral evolution.
We tracked the possible mutational paths between the germline and two broadly neutralising antibodies to show that the exposure history to antigens and the molecular landscape both matter. These results, if generalizable, may explain the molecular basis for the widespread observation that sequential exposure favors greater breadth, and such mechanistic insight will be essential for predicting and eliciting broadly protective immune responses. The methodological advance shows that in-lab evolution can be used to map out the molecular landscape for co-evolution and link genotype to phenotype.
We showed the role of dimensionality of effective space for viral escape and the regime in which influenza operates in terms of cross-reactivity, showing that escape is expected. Linking predictions in phenotypic space to data — this open up the way for further more precise work like this.
