Mapping the functional landscape of T cell receptor repertoires by single-T cell transcriptomics

Abstract

Many experimental and bioinformatics approaches have been developed to characterize the human T cell receptor (TCR) repertoire. However, the unknown functional relevance of TCR profiling significantly hinders unbiased interpretation of the biology of T cells. To address this inadequacy, we developed tessa, a tool to integrate TCRs with gene expression of T cells, in order to estimate the effect that TCRs confer upon the phenotypes of T cells. Tessa leveraged techniques combining single cell RNA-sequencing with TCR-sequencing. We validated tessa and showed its superiority over existing approaches that investigate only the TCR sequences. With tessa, we demonstrated that TCR similarity constrains the phenotypes of T cells to be similar, and dictates a gradient in antigen targeting efficiency of T cell clonotypes with convergent TCRs. We showed this constraint could predict a functional dichotomization of T cells post-immunotherapy treatment, and is weakened in tumor contexts.

A comparative study of multiple instance learning methods for cancer detection using T-cell receptor sequences

“…Under the MIL framework, each sample is considered as a bag consisting of TCR sequences (instances), which are essentially text strings. We embed each TCR sequence into a numeric vector using our previously published Tessa model [38] , which is equipped with a deep learning auto-encoder that converts complex information (strings of amino acids in this case) to numeric values. In short, each amino acid of a TCR sequences is encoded by the five Atchley factors [39] that can fully capture their physicochemical properties.…”

“…A stacked auto-encoder is then applied to the “Atchely matrices” of TCRs to represent the Atchley-factor-encoded TCR sequences by d -dimensional numeric vectors through a decomposition-reconstruction process. Our previous work has systematically established the validity of this approach [38] . By representing each TCR sequence using a numeric vector, we make it convenient for MIL methods to utilize these features, for instance, to calculate distances among instances and/or bags.…”

“…Consisting of amino acid sequences, TCRs are essentially text strings that need to be converted to numeric values before applying MIL methods. In the analysis of TCGA data (Section 5 ), TCRs are converted into numeric vectors by the Briseis encoder [38] . In our simulation, rather than generating TCR sequences, we directly generate numeric values for instances to simplify the process.…”

Antigen-Specific Treg Therapy in Type 1 Diabetes – Challenges and Opportunities

“…Others used traditional statistical approaches (66), or advanced deep learning methods (67)(68)(69)(70)(71)(72)(73) to integrate multiple data sources at once to represent the joint information of all omics-layers. Along these lines, a recent method by Zhang et al jointly integrated TCR and transcriptomic information using Bayesian clustering based on the TCR sequence and gene expression profile (74). Through this method Zhang et al could show that joint TCR and gene expression analysis better separates T-cell specificity and captures the antigen binding efficiency gradient better than TCR-information alone (74).…”

Integrating T-cell receptor and transcriptome for large-scale single-cell immune profiling analysis

“…Due to greater datasets, new methods were proposed utilizing deep learning models such as DeepTCR, which embeds T cells based on their TCR sequence and VDJ gene usage with an Autoencoder (Sidhom et al, 2021). Recently, two novel methods were proposed that additionally include transcriptomic information by Bayesian Clustering (Zhang et al, 2021) or neighborgraph analysis (Schattgen et al, 2020). However, none of the existing methods have addressed learning a joint representation guided by both functional information from TCR sequences and transcriptional profiling of cells to analyze such paired data.…”