Probing T-cell response by sequence-based probabilistic modeling

Abstract

With the increasing ability to use high-throughput next-generation sequencing to quantify the diversity of the human T cell receptor (TCR) repertoire, the ability to use TCR sequences to infer antigen-specificity could greatly aid potential diagnostics and therapeutics. Here, we use a machine-learning approach known as Restricted Boltzmann Machine to develop a sequence-based inference approach to identify antigen-specific TCRs. Our approach combines probabilistic models of TCR sequences with clone abundance information to extract TCR sequence motifs central to an antigen-specific response. We use this model to identify patient personalized TCR motifs that respond to individual tumor and infectious disease antigens, and to accurately discriminate specific from non-specific responses. Furthermore, the hidden structure of the model results in an interpretable representation space where TCRs responding to the same antigen cluster, correctly discriminating the response of TCR to different viral epitopes. The model can therefore be used to identify both experimental and condition specific responding TCRs, of great importance to advancing personalized immunotherapies.

Machine learning identifies T cell receptor repertoire signatures associated with COVID-19 severity

“…Sequence motif studies can be of great clinical significance in broad contexts. Immune cell receptor motif-based investigations are increasingly becoming high-utility, where systematic investigations of specific CDR3 motifs have identified T cell clones associated with specific immune functions like gluten hypersensitivity in celiac disease 41 , hyperinflammation in ankylosing spondylitis 42 , and reactivity to cancer neoepitopes 43 . As longitudinal studies further uncover physiological and immunological effects of long COVID, where patients experience long-term adverse effects from past COVID infection, it is potentially of great interest to compare the TCR repertoire profiles of these patients to the TCR repertoire features identified here.…”

Learning the differences: a transfer-learning approach to predict antigen immunogenicity and T-cell receptor specificity

“…RBM and PWM-based approaches require sequence inputs of fixed length, hence we performed an alignment. Background datasets are aligned to obtain same-length sequences, following the alignment procedures described in [Bravi et al, 2021b] for peptides and [Bravi et al, 2021a] for CDR3 β amino acid sequences. The length of the alignment is set to 9 (9 being the typical length of HLA-I ligands) and to 20 in the case of CDR3 β .…”