Abstract:Understanding the structure of the human T-cell receptor repertoire is a crucial precondition to understand the ability of the immune system to recognize and respond to antigens. T-cells are often compared via the complementarity determining region 3 (CDR3) of their respective T-cell receptor beta chains. Nevertheless, previous studies often simply compared if CDR3beta sequences were equal, while network theory studies were usually limited to several thousand sequences due to the high computational effort of constructing the network. To overcome that hurdle, we introduce the GPU-based algorithm TCR-NET to construct large-scale CDR3beta similarity networks using model-generated and empirical sequence data with up to 800,000 CDR3beta sequences on a normal computer for the first time. Using network analysis methods we study the structural properties of these networks and conclude that (i) the fraction of public TCRs depends on the size of the TCR repertoire, along with the exact (not unified) definition of "public" sequences, (ii) the TCR network is assortative with the average neighbor degree being proportional to the squareroot of the degree of a node and (iii) the repertoire is robust against losses of TCRs. Moreover, we analyze the networks of antigen-specific TCRs for different antigen families and find differing clustering coefficients and assortativities. TCR-NET offers better access to assess large-scale TCR repertoire networks, opening the possibility to quantify their structure and quantitatively distinguish their ability to react to antigens, which we anticipate to become a useful tool in a time of increasingly large amounts of repertoire sequencing data becoming available.

On generative models of T-cell receptor sequences

Statistical inference of the generation probability of T-cell receptors from sequence repertoires

TCR-GPT: Integrating Autoregressive Model and Reinforcement Learning for T-Cell Receptor Repertoires Generation

Predicting Antigen Specificity of Single T Cells Based on TCR CDR 3 Regions

T cell receptor binding prediction: A machine learning revolution

GGNpTCR: A Generative Graph Structure Neural Network for Predicting Immunogenic Peptides for T-cell Immune Response

TCR-TRANSLATE: Conditional Generation of Real Antigen Specific T-cell Receptor Sequences

GRATCR: epitope-specific T cell receptor sequence generation with data-efficient pre-trained models

Deep generative selection models of T and B cell receptor repertoires with soNNia

T-cell receptor structures and predictive models reveal comparable alpha and beta chain structural diversity despite differing genetic complexity

Revealing the hidden sequence distribution of epitope-specific TCR repertoires and its influence on machine learning model performance

Self-supervised learning of T cell receptor sequences exposes core properties for T cell membership

Genesis of the alpha beta T-cell receptor

Population variability in the generation and thymic selection of T-cell repertoires

Scaling Monte-Carlo-Based Inference on Antibody and TCR Repertoires

T-Cell Receptor Cognate Target Prediction Based on Paired α and β Chain Sequence and Structural CDR Loop Similarities

Multi-modal generative modeling for joint analysis of single-cell T cell receptor and gene expression data

Large-scale GPU-based network analysis of the human T-cell receptor repertoire

AIRIVA: A Deep Generative Model of Adaptive Immune Repertoires

Integrating T-cell Receptor and Transcriptome for Large-Scale Single-Cell Immune Profiling Analysis

Restoring data balance via generative models of T-cell receptors for antigen-binding prediction