Analytical code for Pekel and Karcher et al., 2026

Zenodo (CERN European Organization for Nuclear Research)

Abstract

Overview This repository contains the scripts needed to reproduce the analysis and figures from the following manuscript: Meta-analysis reveals a universal microbiome signature for colorectal cancer irrespective of onset age and sequencing method Selin Pekel*, Nicolai Karcher*, Morgan Essex, Fabian Springer, Stefano Romano, Quinten R. Ducarmon, Christian Schudoma, Martin Larralde, Alberto Lupatin, Sebastian Zeissig, Michael Zimmermann, Georg Zeller Requirements This project requires R version 4.3.3 or later. To ensure all required packages are installed, run the following command from the crc-meta-ii/src/directory: Rscript requirements.R This project requires the SIAMCAT R package, specifically version 2.7.2, which includes critical updates necessary for compatibility with the current workflow. Using an older version may result in errors or unexpected behavior. For detailed usage instructions and documentation, please visit the SIAMCAT website. In addition, SHAP value calculations in this project were performed using classifiers built with the mlr3 package (version 0.20.2), so please ensure that this version is installed to avoid compatibility issues. Workflow Overview 1. Model Training (Random Forest) Machine learning models (Unified CRC, EO-CRC, LO-CRC, etc.) are trained using scripts in the models/directory. Output files from model training are saved to data/results/. To run LOSO models and evaluate both CRC and AD samples, use the scripts: input.LOSO.rf.models.sh 04.run.Train.LOSO.rf.models.sh\ These scripts must be submitted via SLURM. To obtain study-to-study transfer (SST) evaluation AUCs, run: 05.Train.SCV.SST.rf.models.R Outputs from both LOSO and SST evaluations are stored in: data/results/scv.loso/ 2. Statistical Analysis & Linear Mixed Models All statistical analysis scripts are located in src/analysis/. Example scripts: 02.lmm.16S.WGS.R: LMMs comparing CRC vs. CTR in 16S and WGS profiles 03.lmm.EO.LO.R: LMMs for EO-CRC vs. EO-CTR or LO-CRC vs. LO-CTR comparisons 3. Model Evaluation & SHAP Analysis SHAP interpretation scripts are located in src/analysis/ and are submitted via SLURM (04*.sh). Median SHAP values are computed and saved under: data/results/shap.analysis/ 4. Figure Generation All manuscript figures are generated using scripts in src/plotting/, e.g.: Figure1.R, Figure2.R, ..., Figure6.R

Authors 3

1. Selin Pekel Aachen

   Leiden University Medical Center

   Affiliation as printed

   Leiden University Medical Center

2. Nicolai Karcher

   European Molecular Biology Laboratory · European Molecular Biology Laboratory

   Affiliation as printed

   European Molecular Biology Laboratory

3. Georg Zeller Aachen

   European Molecular Biology Laboratory · European Molecular Biology Laboratory · Leiden University Medical Center

   Affiliation as printed

   Leiden University Medical Center

   European Molecular Biology Laboratory