The authors present BRIDGE-AD, an interpretable network medicine framework designed to convert heterogeneous, multimodal evidence into a unified, disease-specific gene representation for prioritising effectors in Alzheimer's disease (AD). The stated aim is to harness the growing landscape of AD datasets to systematically discover genes likely to drive disease biology and to enable traceable, mechanistic hypothesis generation.
BRIDGE-AD emphasises interpretability and integrative modelling across multiple evidence layers rather than relying on single-modality or pretrained embeddings alone. The framework is intended both to recover known AD-associated genes and to nominate new candidate effectors for further study.
BRIDGE-AD integrates more than 30 datasets and curated resources encompassing diverse evidence types. The integration spans:
The manuscript reports that these layers were combined into a single, disease-specific gene representation; specific dataset names and the precise integration algorithms are referenced in the preprint and accompanying code repository. The resource generated is genome-wide in scope, enabling prioritisation across the entire gene set.
According to the authors, BRIDGE-AD outperformed recently published pretrained and modality-specific gene embeddings in recovering AD-associated genes. The claim is based on benchmarking performed by the authors comparing BRIDGE-AD representations to alternative embeddings. Details of the benchmark metrics, datasets used for evaluation, and exact performance numbers are reported in the source preprint and accompanying materials.
Using the integrated representations, BRIDGE-AD produced a genome-wide resource of candidate AD effectors. The prioritised set included both previously established AD-associated genes and newly nominated candidates. The authors grouped these genes into 19 functional clusters, which they interpret as revealing a global molecular landscape of AD biology. These clusters reflect convergent biological programmes implicated by the integrated evidence layers.
The clusters and their component genes serve as a structured map to contextualise candidate effectors and to support mechanistic hypothesis generation across pathways and cell compartments relevant to AD.
BRIDGE-AD-supported analyses highlighted an SPP1-centred cross-compartment hypothesis. The authors describe SPP1 as a central node in a cross-compartment molecular hypothesis for AD based on the integrated evidence, suggesting coordinated roles across cell types and biological processes. The preprint details the evidence supporting this hypothesis within the integrated network outputs.
Among the newly prioritised candidates, the framework nominated SCARB2 as a poorly characterised gene warranting functional follow-up. The authors report experimental validation indicating that SCARB2 perturbs lysosomal, lipid-handling, and autophagic programmes in microglia.
Additionally, the preprint notes observations of disrupted SCARB2 glycosylation in AD, which the authors interpret as implicating altered SCARB2 processing and function in the disease context. These functional findings are presented as case evidence of how BRIDGE-AD can guide downstream experimental investigation of nominated effectors.
To promote transparency and facilitate use of the integrated resource, the authors provide an interactive website at explore-bridgead.com that enables users to trace the curated evidence supporting gene prioritisation and to generate mechanistic hypotheses. A code repository is available at https://github.com/Greta-B/BRIDGE-AD, where scripts, data links, and implementation details are provided according to the preprint.
Users interested in reproducing results or exploring specific genes can consult these resources for the curated datasets, integration procedures, and evidence layers underlying the BRIDGE-AD gene representations.
The work was funded by multiple sources reported in the preprint, including the National Institute on Aging (NIH grants R01 AG072291 and R01 AG079307), the NIHR Cambridge Biomedical Research Centre, the National Research Foundation of Korea, the Brain Pool Plus Fellowship Program, and CIRM training support, among others. The preprint notes specific grant identifiers in its funding declaration.
A competing interest disclosure states that Namshik Han is co-founder and CTO of CardiaTec Bio and co-founder of KURE.ai and serves on an external scientific advisory board; the authors state these affiliations are unrelated to the manuscript. The remaining authors declared no competing interests.
The preprint is available on bioRxiv (doi: https://doi.org/10.64898/2026.09.14.750802) and was posted September 20, 2026. The manuscript is shared under a CC-BY-NC-ND 4.0 International license.
BRIDGE-AD provides an interpretable, integrative approach that the authors propose will aid discovery of disease effectors and foster mechanistic follow-up in AD research. By combining many evidence layers and offering traceable links to curated sources via an interactive website and public code repository, the resource is positioned to support experimental prioritisation, validation efforts, and hypothesis-driven studies of candidate AD effectors.
Researchers interested in using BRIDGE-AD should consult the preprint and the provided website and GitHub repository for methodological details, dataset provenance, and instructions for exploring gene-level evidence. The preprint presents both the computational resource and example experimental follow-up (SCARB2) to illustrate how integrated prioritisation can be translated into functional investigation.