A Falsifiable Framework for Testing Neutralityin T-Cell Receptor Repertoire Databases

BackgroundT-cell receptor (TCR) repertoire databases aggregate antigen-specific sequences from hundreds of studies, exhibiting heavy-tailed occurrence distributions commonly interpreted as signatures of selection or criticality. However, distinguishing genuine non-neutral dynamics from neutral drift with realistic biological constraints requires explicit computational null models--currently absent from repertoire immunology. MethodsWe developed a biologically calibrated neutral null model incorporating thymic output, negative selection, and peripheral homeostasis with parameters independently derived from immunology literature (Robins et al. 2009, naive repertoires). The model was validated against empirical repertoire statistics before generating predictions. We compared neutral predictions to occurrence patterns from 24,847 TCR-epitope combinations in VDJdb across four viral targets using multi-observable testing (power-law exponent, Shannon entropy, public clonotype fraction) with Bonferroni-corrected thresholds. ResultsThe neutral null model successfully reproduced three independent benchmarks. Public clonotype fraction in VDJdb (3.10%) significantly exceeded neutral predictions (1.47%{+/-}0.29%, 100th percentile, p < 0.001, Cohens d = 5.62), inconsistent with neutrality at stringent thresholds (Bonferroni ' = 0.0167). In contrast, power-law exponent ( = 2.450 vs. 2.391 {+/-} 0.177, p = 0.739, d = 0.33) and Shannon entropy (H = 12.10 vs. 12.34 {+/-} 1.17 bits, p = 0.838, d = -0.21) showed negligible deviations. This dissociation--public fraction deviates while diversity metrics remain neutral-consistent--is consistent with but does not prove selective enrichment for cross-individual TCR convergence. Supplementary analyses confirmed public enrichment is temporally stable (2009-2024, no significant trend p > 0.40) and universal across viral pathogens (CMV, EBV, Influenza, SARS-CoV-2, ANOVA p > 0.68), constraining plausible curation bias mechanisms. ConclusionsWe present a rigorous falsificationist framework for testing neutrality in TCR repertoires via pre-validated computational nulls, multi-observable comparisons, and honest power reporting. Application to VDJdb reveals public clonotype patterns suggestive of non-neutral processes (functional selection or curation bias), establishing testable hypotheses for experimental follow-up. The framework--emphasizing independent validation, pre-specified decision criteria, and effect size quantification--generalizes to antibody repertoires, microbiomes, and evolutionary systems requiring mechanistic discrimination between drift and selection. Author SummaryDetermining whether T-cell receptor occurrence patterns arise from random drift or functional selection is fundamental to understanding adaptive immunity, yet current approaches rely on descriptive statistics rather than rigorous hypothesis testing. We developed the first falsifiable framework for testing neutrality in TCR databases using a biologically realistic computational null model validated against independent empirical benchmarks. Applying this framework to VDJdb, we find that public clonotype frequencies (sequences appearing across many individuals) deviate significantly from neutral predictions, while overall repertoire diversity remains consistent with neutrality. This pattern suggests--but does not definitively prove--that selection or curation bias operates specifically on cross-individual TCR convergence. Our contribution is methodological: we demonstrate how to test mechanistic hypotheses rigorously rather than describe patterns phenomenologically. The framework is immediately applicable to ongoing debates in repertoire immunology and generalizes to other biological systems exhibiting heavy-tailed distributions.