Comparing T cell receptor repertoires using optimal transport

Abstract

The complexity of entire T cell receptor (TCR) repertoires makes their comparison a difficult but important task. Current methods of TCR repertoire comparison can incur a high loss of distributional information by considering overly simplistic sequence- or repertoire-level characteristics. Optimal transport methods form a suitable approach for such comparison given some distance or metric between values in the sample space, with appealing theoretical and computational properties. In this paper we introduce a nonparametric approach to comparing empirical TCR repertoires that applies the Sinkhorn distance, a fast, contemporary optimal transport method, and a recently-created distance between TCRs called TCRdist. We show that our methods identify meaningful differences between samples from distinct TCR distributions for several case studies, and compete with more complicated methods despite minimal modeling assumptions and a simpler pipeline.

Recruitment of naive CD4+ T cells by the recombinant zoster vaccine correlates with persistent immunity

“…To capture the full potential CD4 + T cell responses after RZV or ZVL, we used a computational sequence similarity algorithm (TCRdist) ( 22 , 23 ) that identified closely related TRB sequences in either memory or naive prevaccination TRB repertoires. Using a conservative definition of amino acid similarity (up to 4 chemically similar amino acid changes in the CDR3 region or a single chemically dissimilar amino acid change or deletion) ( 22 – 24 ) and normalized to the input number of cells, we found that RZV recipients had significantly more clonotypes in their predicted gE-reactive swarms than ZVL recipients. This was true for sequence-related TRB identified by TCRdist for both peak (median 285 per RZV recipient versus 99 per ZVL recipient, P = 0.02; Figure 2B ) and lasting (53.3 versus 5, P = 0.005; Figure 2B ) CD4 + clonotypes.…”

Predicting TCR sequences for unseen antigen epitopes using structural and sequence features

“…A variety of machine learning methods have been applied, ranging from clustering-based approaches such as TCRdist (2022) [ 32 ], GLIPH (2017) [ 33 ] and TCRMatch (2021) [ 34 ] to random forest algorithms such as epiTCR (2023) [ 26 ] to address these challenges. The influx of data has inspired deep learning techniques, including gated recurrent unit (GRU) and transformer models, which have been adapted from their success in natural language processing to improve TCR–epitope binding predictions with models such as ERGO (2020) [ 35 ], ImRex (2021) [ 36 ], TITAN (2021) [ 25 ], DeepTCR (2021) [ 37 ], TEINet (2023) [ 27 ], PanPep (2023) [ 38 ] and TEIM-Seq (2023) [ 39 ].…”

CoV2-TCR: A web server for screening TCR CDR3 from TCR immune repertoire of COVID-19 patients and their recognized SARS-CoV-2 epitopes

“…imRex [32] , ERGO-LSTM [33] , and ERGO-AE [33] . In addition, comparing with the TCRmatch [40] and TCRdist [41] constructed by the similarity-based or distance-based approaches, the CoV2-TCR was established by convolution neural network, whose advantage lies in its capacity of feature extraction. NetTCR-2.0 was trained to predict interactions between TCRs and their cognate HLA-A*02:01-restricted peptides [42] .…”