Profiling of AIRR generative models

Work in progress 🚧

This project explores how to evaluate generative models for Adaptive Immune Receptor Repertoire (AIRR) data and provides profiling analyses of popular, existing models in the field.

We aim to better understand the behaviour and output of these models by applying repertoire-level statistics, distributional analyses, and other comparisons to empirical immune receptor datasets.

Goals

• Explore and assess evaluation methods for generative AIRR models
• Profile outputs of commonly used models (e.g., VAE, LSTM, soNNia)
• Compare generated repertoires to real-world immune data
• Identify strengths, biases, and limitations across models

What This Is

• An exploratory research project, not a framework or software package
• A collection of scripts, notebooks, and results to support analysis and reproducibility
• A work in progress — structure, methods, and ideas are actively evolving