ReplicateFest, a software framework for T-cell receptor sequencing, recovered all nine FEST-expanded clonotypes previously reported in an HIV-1 dataset. It also recovered four of five previously reported cross-reactive clonotypes, while flagging one additional cross-reactive clonotype. The comparison shows that different calling rules can produce different answers. The authors attribute the mismatch to different methods for calling FEST-expanded clonotypes.
A split route for experiments with and without repeats
The study centers on FEST-based T-cell receptor sequencing data. T cells are cultured in parallel under one condition per peptide, alongside a no-peptide control, and the design allows biological or technical replicates. In this analysis, an expanded clonotype is a receptor sequence that appears more often under a peptide condition than in the reference. ReplicateFest is built to analyze those experiments with or without replicates.
Before testing, the import step keeps productive clonotypes, combines nucleotide sequences that translate to the same amino-acid CDR3, and removes clonotypes with low template counts.
For non-replicate data, the framework uses Fisher's exact test. For replicated experiments, it uses negative-binomial modeling across conditions, a count-based approach for repeated samples. It returns adjusted p-values and odds ratios for the comparisons.
The default FEST-expanded filter requires a false-discovery rate below 0.05 and a positive odds ratio against the reference. FEST-positive clonotypes are a subset that must be expanded against all the other tested conditions, not only the no-peptide control.
The HIV-1 reanalysis mostly matched earlier calls
The HIV-1 analysis excluded samples with fewer than 5,000 productive reads and retained 79 samples covering 25 peptide conditions plus a no-peptide reference. ReplicateFest v0.1.0 ran with cross-reactivity enabled, using a false-discovery-rate threshold of 0.01 and an odds-ratio threshold of 5.
Under those settings, the reanalysis recovered all nine reported FEST-expanded clonotypes. It recovered four of five reported cross-reactive clonotypes and added one cross-reactive call that Chen et al. had not reported. The authors attribute the difference to the methods used to decide which clonotypes qualify as FEST-expanded.
That agreement does not establish antigen specificity. The authors say the simulations and the HIV-1 analysis did not examine specificity.
Small and modest expansions were hardest to catch
To test behavior in a controlled setting, the study created artificial repertoires. In experiments with and without replicates, the software identified both FEST-expanded and FEST-positive clonotypes. Each experiment contained five peptide conditions, one expanded clonotype per condition, and one no-expansion condition.
A broader simulation crossed three starting-frequency levels with three expansion factors: twofold, fourfold and eightfold. The starting levels corresponded to the 50th, 75th and 95th percentiles of the empirical mean. The evaluation used 10 datasets with 90 expanded clonotypes in each.
Detection was near zero for twofold expansion. At fourfold expansion, it ranged from 0.42, with reported variability of 0.10, at the lowest starting-frequency level to 0.92, with reported variability of 0.10, at the highest. At eightfold expansion, detection was at least 0.98 across starting frequencies, with no variability reported at the highest level.
The pattern was clear: the software was far less likely to detect a small expansion from a low starting frequency than a large expansion from a common starting frequency in these simulations. That is a warning for sparse or modest signals, which may be missed even when a clone is present.
A framework, not a biological verdict
The study presents replicateFest as a framework for analyzing these data, but the validation is narrower than a biological experiment. The simulations did not model expansion kinetics, competition between clones, changes in the repertoire, sampling effects or sequencing-depth variation. Performance may also be weaker for very sparse clonotypes, low sequencing depth or experiments with few replicates.
The framework is available as an R package and Shiny application under the Apache-2.0 license, with the package requiring R 4.0 or newer. It provides one implementation for the replicate-aware and non-replicate analysis routes described in the study.
Paper data and sources
Original title: replicateFest: an R package and shiny app for analysis of T cell receptor repertoire data from the functional expansion
Authors: Danilova L, Favorov A, Smith K, Cope L
Journal/Repository: Frontiers in immunology
Status: Peer-reviewed
First online: 2026-10-19
DOI: Not available
Original paper · Full text