Quantifying the Recoverability of V and J Genes from TCR CDR3 Sequences Using Generative Repertoire Models

A study investigates the extent to which the variable (V) and joining (J) gene identity of a T-cell receptor's third complementarity-determining region (CDR3) sequence can be determined. Researchers analyze 118,096 pooled human rearrangements, computing posterior distributions of candidate genes under two models: one considering recombination and another accounting for post-selection.

They measure recoverability using conditional entropy, the size of the candidate gene list, the fraction of sequences allowing a high-confidence single-gene determination, and the gene-by-gene confusion structure. The findings reveal that the J gene is almost entirely determined by the CDR3 in both beta and alpha chains. However, the V gene's recoverability is incomplete, functioning more as a group than an individual gene.

Factors such as junctional trimming, non-templated insertion, and the loss of synonymous codon information contribute to this group behavior, with gene sets differing significantly from germline family nomenclature. Applying selection refines the V gene posterior distribution but alters the most probable gene, a phenomenon specific to beta chains.

Selection also reshuffles which V gene is most likely, particularly in beta chains, contrasting with only a mild reweighting in alpha chains. The study's methods and results are consistent across two additional tumor cohorts. The researchers conclude that V identity should be reported as an emergent, system-level property of the repertoire, represented as a calibrated group rather than a single gene.

They provide a computational tool that outputs a set of candidate genes with associated confidence values based on a CDR3 sequence.