A deep-learning–informed prior and Bayesian model for differential AP-MS interactome analysis

This record archives the software and the self-contained reproducibility package accompanying the methods paper "A deep-learning–informed prior and Bayesian model for differential AP-MS interactome analysis" (M. Seefelder, Department of Gene Therapy, Ulm University). What BayesInteractomics is. Affinity-purification mass spectrometry (AP-MS) maps a bait protein's partners, but every purification also captures abundant non-specific background that masks genuine interactors. BayesInteractomics is an open-source Julia framework that combines machine learning with Bayesian statistics: a neural network trained on protein structures predicts direct binding, a calibrated meta-learner turns this into an informed prior, and the prior guides a Bayesian copula-mixture model that integrates three AP-MS evidence streams — enrichment, co-abundance, and detection reproducibility. Each candidate receives an interaction probability at a controlled false-discovery rate, and the method also resolves which interactions are gained or lost between two conditions. What this archive contains. Two things, versioned together: The exact source snapshot of the method documented in the paper, BayesInteractomics.jl v1.2.1 (MIT licence; development home: github.com/ma-seefelder/BayesInteractomics). A reproducibility package bundling every script, dataset, and committed intermediate result needed to regenerate the manuscript's quantitative figures and tables, plus per-figure Source Data workbooks (one .xlsx per statistical figure) and SHA-256 checksums. Two tiers of reproducibility. Tier A (offline, seconds): every data-bearing figure recompiles from its committed intermediate CSV with a single Typst command (pinned to Typst 0.14). No Julia, R, or comparator installation required — this is the checkable guarantee. Tier B (heavy, optional): regenerating the intermediate data from raw mass-spectrometry inputs runs the full pipeline (Julia 1.12, SAINTexpress 3.6.3, R 4.5.2 with CompPASS/SMAD and MiST, Python 3.14). Exact commands and seeds are documented per component; the synthetic generator's master seed is 20260528. Contents. Synthetic AP-MS data generator; comparator wrappers (SAINTexpress, CompPASS, MiST); head-to-head benchmark and architectural-ablation drivers; Typst figure modules with committed plot data; per-figure Source Data; a machine-readable figure_manifest.csv mapping each figure to its component, data, and regeneration command; and a verify_package.sh harness (checksum, provenance, and figure-regeneration gates). Real datasets re-analysed (raw data available from PRIDE, not redistributed here): OpenCell (PXD024909), Greco et al. 2022 (PXD025510), Sap et al. 2021 (PXD024254), and Hein et al. 2015 (PXD002815). The curated interaction reference combines CORUM and IntAct. Citation. If you use this software or package, please cite the accompanying methods paper and this Zenodo record (DOI: 10.5281/zenodo.21217041).