Pathogen-derived and host-reactive peptide signatures of HIV immune response

Dual-layer serological encoding of HIV infection by peptide-array profiling — data and analysis code

This Zenodo record contains the primary peptide-array data, cohort metadata, and complete Python analysis code supporting the manuscript "Dual-layer serological encoding of HIV infection by peptide-array profiling" (Schmidt et al., 2026).

Experimental design. Serum samples from 105 individuals (35 HIV-positive, 70 HIV-negative) collected in South Africa, Peru, and Viet Nam were profiled on a custom in-situ-synthesised peptide microarray comprising two parallel libraries: (i) a pathogen-derived panel of 49 HIV-1 twelve-mer peptides covering the gp41 envelope and p24 Gag proteins, and (ii) a host-resembling library of 2,087 ten-mer peptides selected by compositional similarity to human proteomic fragments. IgG binding was quantified by fluorescence scanning after secondary-antibody staining. Under strict nested 5×5 cross-validation, the pathogen-derived panel classified HIV status with AUC 0.891; the host-resembling library alone reached AUC 0.805; probability-level meta-stacking yielded AUC 0.897. HIV-positive sera additionally showed significantly reduced Shannon entropy (p = 0.012) and elevated Gini inequality (r = 0.376) in the host-peptide binding distribution, consistent with antibody-repertoire remodelling under chronic infection.

Contents of this deposit.

• data/results_HIV_final.pkl — per-sample HIV-panel fluorescence data (Python pickle; 105 samples × 49 peptides, median intensity and replicate SD)
• data/results_RRL_final.pkl — per-sample host-resembling-library data (105 samples × 2,087 peptides)
• data/sample_metadata.csv — cohort annotation: HIV status, age, sex, country, tuberculosis co-infection status, array-level QC notes
• data/hiv_peptides.csv — 49-peptide annotation with LANL source strain, HXB2 position, sequence, and category (template vs scrambled control)
• notebooks/1_HIV_RF.ipynb, 2_RRL_RF.ipynb, 3_Covariance_and_Meta.ipynb — Jupyter notebooks implementing the nested-cross-validation Random Forest classifiers, probability-level meta-model integration, and per-sample distributional metrics (Shannon entropy, Gini index, signal variance, top-peptide dominance)
• README.md — deposit guide and reproduction instructions
• environment.yml — conda environment specification for re-running the notebooks (Python 3.11, scikit-learn 1.2)
• LICENSE_CODE.md (MIT) and LICENSE_DATA.md (CC-BY-4.0) — licensing terms

How to reproduce. Create the conda environment from environment.yml, launch JupyterLab, and run the three notebooks in order. Reported AUCs reproduce within rounding under fixed random seeds.

Known structural features of the cohort. HIV status is not balanced across countries (Peru 1/37 HIV+, South Africa 24/37, Viet Nam 10/31); leave-one-country-out sensitivity analyses are therefore advisable when re-analysing the data. Clinical metadata beyond HIV status, age, sex, country, and tuberculosis status are not publicly available owing to the terms of the FIND Foundation Biobank data transfer agreement and may be requested from the corresponding authors subject to institutional review.

Citation. When re-using any material from this deposit, please cite the associated publication and this Zenodo record. Code is released under the MIT licence; data under Creative Commons Attribution 4.0 International (CC-BY-4.0).

Contact. Alexander Nesterov-Mueller (alexander.nesterov-mueller@kit.edu) · Alex Dulovic (alex.dulovic@nmi.de)