Longitudinal TCR clonotype trajectories and transition datasets across detectability thresholds

This dataset contains longitudinal T cell receptor (TCR) clonotype trajectories and derived temporal transition tables generated within the ClonoDynamics pipeline tcr-dynamics-noise-pipeline. In particular, trajectory tables were produced with trajectory_builder.py, and transition tables were generated from these trajectories with build_transition_patterns.py. The pipeline is available in the ClonoDynamics GitHub repository: https://github.com/ClonoDynamics/tcr-dynamics-noise-pipeline. For each detectability threshold (α), the dataset provides: Trajectory tables describing clonotype abundance over time Transition tables describing pairwise temporal changes in log-frequency These representations enable reproducible inference of clonotype dynamics, including drift, diffusion, and stochastic modeling under different detectability regimes. 📊 Data structure 1. Trajectory tables (trajectories_long.csv) Each row represents a clonotype at a given time: subject — subject identifier time — sampling time point aaSeqCDR3 — clonotype identifier freq — clonotype frequency log_freq — natural logarithm of frequency observable — Boolean detectability indicator 2. Transition tables (transitions_all.csv) Each row represents a temporal transition between two timepoints for the same clonotype. Core variables subject — subject identifier aaSeqCDR3 — clonotype identifier t0, t1 — initial and final timepoints dt = t1 − t0 — time lag x0, x1 — log-frequency at t0t0 and t1t1 dx = x1 − x0 — log-change in abundance Δx=log⁡f(t1)−log⁡f(t0)Δx=logf(t1)−logf(t0) Detectability information obs0, obs1 — detectability at endpoints (T/F) obs_class — transition class: TT: detectable → detectable TF: detectable → non-detectable FT: non-detectable → detectable FF: non-detectable → non-detectable Transition structure pattern — transition type (all = all valid pairs) n_steps — number of time steps (equal to dt) Weights To support balanced statistical inference, each transition is associated with weight components: w_clone — inverse number of transitions for that clonotype w_dt — inverse frequency of transitions with the same time lag dtdt w_class — class-specific weight (TT, TF, FT, FF) w — total weight: w=w_clone⋅w_dt⋅w_class bayes_weight — Bayesian transition-confidence weight derived from posterior uncertainty in the inferred displacement. Transitions with lower posterior variance contribute more strongly to downstream drift and diffusion inference, whereas transitions with high posterior uncertainty are automatically downweighted.