DextraDemixer Reproducibility Package

This repository contains the code, notebooks, and selected precomputed results used to reproduce analyses from the manuscript DextraDemixer enables accurate identification of antigen-specific T cells from pMHC multimer experiments.

If you want to use a clean version for your own project, please use https://github.com/SchubertLab/DextraDemixer

The reproducibility material is organized by manuscript figure:

• experiments/synthetic_benchmark: Figure 2 and supplementary synthetic benchmark analyses.
• experiments/Gemuend_CMV: Figure 3, based on the CMV spike-in dataset from Gemuend et al.
• experiments/Kocher_SARS-CoV-2: Figure 4, based on the SARS-CoV-2 dataset from Kocher et al.

The Python package code is in dextrademixer/.

Installation

Create the recommended Conda environment with versions for the most important packages only:

conda env create -f environment_minimal.yaml
conda activate dextrademixer

If you need the exact exported environment used for the original analyses, use:

conda env create -f environment_full.yaml
conda activate dextrademixer

The full environment may not resolve on all systems. Use environment_minimal.yaml unless you specifically need the fully pinned environment.

Reproducing the Figures

Run the notebooks from the corresponding experiment directories. Several plotting notebooks can be run directly from the precomputed result files included in this repository, notebooks that rerun full model fitting or use external data may require manually downloaded source data.

Figure 2: Simulated Data

Directory: experiments/synthetic_benchmark

The synthetic benchmark was run with Snakemake. The workflow generates simulated datasets and runs the benchmarked methods. See snakefile_benchmark.smk and slurm_benchmark.sh for the full benchmark setup.

The simulated datasets are deposited at Zenodo: https://doi.org/10.5281/zenodo.20759653

Notebooks:

• Figure2_synthetic_benchmark.ipynb: recreates the main Figure 2 panels from result files included in this repository.
• SuppFig2_Estimate_sim_params.ipynb: estimates simulation parameters from real 10x Genomics data; this requires manual data download as described in the notebook.
• SuppFig3_Examples_of_simulated_datasets.ipynb: plots examples of simulated datasets; this requires downloading the simulated datasets from Zenodo.

Figure 3: CMV Spike-In Dataset From Gemuend et al.

Directory: experiments/Gemuend_CMV

Notebooks:

• Figure3a_run_benchmark.ipynb: runs DextraDemixer, BEAM, ICON, and ITRAP on the experimental benchmark. This requires downloading the original data and preprocessing it with the original script from https://gitlab.dzne.de/ag-beyer/gemuend_cmv_2025, using their Part2_CMV-Dex_Multiome.ipynb notebook.
• Figure3b_plotting.ipynb: plots the benchmark results. With the included result file, some panels can be reproduced without rerunning Figure3a_run_benchmark.ipynb.

Figure 4: SARS-CoV-2 Dataset From Kocher et al.

Directory: experiments/Kocher_SARS-CoV-2

Notebooks:

• Figure4a_run_dextrademixer.ipynb: fits DextraDemixer on the Kocher et al. dataset. This requires downloading the raw data from the original source as described in the notebook.
• Figure4b_analysis.ipynb: analyzes DextraDemixer predictions. The required predictions are included in this repository, so this can be run without rerunning Figure4a_run_dextrademixer.ipynb.
• Figure4c_plotting.ipynb: plots the final Figure 4 results. The required precomputed results are included in this repository, so this can be run without rerunning Figure4b_analysis.ipynb.

Additional precomputed files for this analysis are stored under experiments/Kocher_SARS-CoV-2/results/.

Citation

If you use this repository, please cite the manuscript:

DextraDemixer enables accurate identification of antigen-specific T cells from pMHC multimer experiments

License

This project is distributed under the BSD 3-Clause License. See LICENSE for details.