This repository contains the manuscript-specific code, notebooks, analysis scripts, figure-generation notebooks, and reproduction materials associated with the PepCL peptide–MHC class I continual learning study.

This repository is intended primarily as a manuscript reproduction repository. It is not structured as a pip-installable Python package. Packaged, reusable versions of the main software components are available separately:

• MHCPrime: https://github.com/ntranoslab/MHCPrime
• PepCL: https://github.com/ntranoslab/PepCL

Users who want to run the packaged base model or PepCL update workflow directly should use the MHCPrime and PepCL repositories linked above. This manuscript repository contains the broader analysis and experiment code used to generate the results reported in the manuscript.

CLSBN26/
├── Analysis/
├── src/
├── Figures/
├── LICENSE
└── README.md

The Analysis/ directory contains scripts and notebooks used for manuscript analyses, including data collection, data processing, model training, PepCL updating experiments, benchmarking, interpretability analyses, and other experiments discussed in the manuscript.

The src/ directory contains project code used by the manuscript analyses, including model definitions, training utilities, PepCL-related code, metric functions, preprocessing utilities, and analysis helpers.

For general use, the packaged versions of the model and PepCL framework should be accessed through the standalone MHCPrime and PepCL repositories:

• https://github.com/ntranoslab/MHCPrime
• https://github.com/ntranoslab/PepCL

The Figures/ directory contains notebooks and scripts used to generate manuscript figures and supplementary figures. It also contains raw figure outputs and associated figure-level data where applicable.

Two large directories are required for full manuscript reproduction but are not included directly in this GitHub repository:

Data/
Model_Checkpoints/

These folders can be downloaded from the CLSBN26 data and model checkpoint folder.

The Data/ directory is required for figure generation and manuscript analyses. The Model_Checkpoints/ directory is required for analyses that load trained or updated model checkpoints.

After downloading, place both folders at the top level of the repository. The expected top-level layout for full reproduction is therefore:

CLSBN26/
├── Analysis/
├── src/
├── Figures/
├── Data/                 # must be downloaded
├── Model_Checkpoints/    # must be downloaded
├── LICENSE
└── README.md

This repository is intended for reproducing manuscript analyses rather than as the primary user-facing software interface.

For most users:

1. Use MHCPrime to load and run the packaged base peptide-MHC model.
2. Use PepCL to run packaged continual learning updates and quickstart workflows.
3. Use this manuscript repository to inspect or reproduce the analyses, training runs, update experiments, and figure-generation workflows described in the manuscript.

The original analyses were developed in a Python/Conda environment with GPU-enabled PyTorch. All training and updating experiments were performed and designed for a single NVIDIA H100.

--