Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 75 of the 75 references it cites were found in a registry or an index.

Effective integration of multi-omics with prior knowledge to identify biomarkers via explainable graph neural networks

Open access
Tripathy, Rohit K., Frohock, Zachary, Wang, Hong, Cary, Gregory A., Keegan, Stephen, Carter, Gregory W., Li, Yi · 2025 · npj Systems Biology and Applications

Abstract

The rapid growth of multi-omics datasets and the wealth of biological knowledge necessitates the development of effective methods for their integration. Such methods are essential for building predictive models and identifying drug targets based on a limited number of samples. We propose a framework called GNNRAI for the supervised integration of multi-omics data with biological priors represented as knowledge graphs. Our framework leverages graph neural networks (GNNs) to model the correlation structures among features from high-dimensional ‘omics data, which reduces the effective dimensions in data and enables us to analyze thousands of genes simultaneously using hundreds of samples. Furthermore, our framework incorporates explainability methods to elucidate informative biomarkers. We apply our framework to Alzheimer’s disease (AD) multi-omics data, showing that the integration of transcriptomics and proteomics data with prior AD knowledge is effective, improving the prediction accuracy of AD status over single-omics analyses and highlighting both known and novel AD-predictive biomarkers.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41540-025-00519-9

Cites

75 of 75 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Schematic of our end-to-end integrative model GNNRAI. Data from individual ‘omics modalities are processed in their respective GNN feature extractors to produce low-dimensional embeddings (z1 and z2), which are then aligned and integrated through a set transformer. Samples with incomplete multi-omics measures have their embeddings processed through separate MLP (multi-layer perceptron) classifiers to produce modality-specific predictions of the target. Some elements of the figure were rendered using functions from Matplotlib in Python, while the entire figure was created in PowerPoint.
Fig. 1 Schematic of our end-to-end integrative model GNNRAI. Data from individual ‘omics modalities are processed in their respective GNN feature extractors to produce low-dimensional embeddings (z1 and z2), which are then aligned and integrated through a set transformer. Samples with incomplete multi-omics measures have their embeddings processed through separate MLP (multi-layer perceptron) classifiers to produce modality-specific predictions of the target. Some elements of the figure were rendered using functions from Matplotlib in Python, while the entire figure was created in PowerPoint.CC BY · PubMed Central
Fig. 2 Three-fold cross-validation performance of our proposed integrative model (red) compared to that of the benchmark MOGONET model (cyan) on the set of common samples having both proteomics and transcriptomics measurements. The performance of the integrative models is also compared to unimodal GNN transcriptomics (blue) and proteomics (green) models.
Fig. 2 Three-fold cross-validation performance of our proposed integrative model (red) compared to that of the benchmark MOGONET model (cyan) on the set of common samples having both proteomics and transcriptomics measurements. The performance of the integrative models is also compared to unimodal GNN transcriptomics (blue) and proteomics (green) models.CC BY · PubMed Central
Fig. 3 Three-fold cross-validation performance of the integrative, multimodal model (red and cyan) compared to that of unimodal GNN models (blue and green) trained on the incomplete multi-omics datasets for 16 AD biodomains. The performance of the multimodal model is calculated on two sets of validation samples – the set of all validation samples with transcriptomics and that with proteomics measurements.
Fig. 3 Three-fold cross-validation performance of the integrative, multimodal model (red and cyan) compared to that of unimodal GNN models (blue and green) trained on the incomplete multi-omics datasets for 16 AD biodomains. The performance of the multimodal model is calculated on two sets of validation samples – the set of all validation samples with transcriptomics and that with proteomics measurements.CC BY · PubMed Central
Fig. 4 Predictive performance of applying unimodal transcriptomics model trained on ROSMAP DLPFC training dataset to validate transcriptomics samples from various cohorts and brain tissues.
Fig. 4 Predictive performance of applying unimodal transcriptomics model trained on ROSMAP DLPFC training dataset to validate transcriptomics samples from various cohorts and brain tissues.CC BY · PubMed Central
Fig. 5 Predictive performance of applying unimodal and multimodal models trained on ROSMAP DLPFC training dataset to samples from ROSMAP DLPFC validation and MSBB PHG validation datasets. Prot protein, Int integrative.
Fig. 5 Predictive performance of applying unimodal and multimodal models trained on ROSMAP DLPFC training dataset to samples from ROSMAP DLPFC validation and MSBB PHG validation datasets. Prot protein, Int integrative.CC BY · PubMed Central
Fig. 6 Interactions between biodomains identified by Integrated Hessians analysis on a multi-modal model trained on samples with complete multi-omics. The nodes represent biodomains, and the edges represent interactions, with the edge annotations being the rank of the interaction. The biodomain names based on the 3-letter abbreviations are: apoptosis (apo), lipid metabolism (lip), synapse (syn), endolysosome (end), metal binding (met), cell cycle (cel), mitochondrial metabolism (mit), myelination (mye), vasculature (vas), proteostasis (pro) and APP metabolism (app). The figure was rendered using functions from Matplotlib in Python.
Fig. 6 Interactions between biodomains identified by Integrated Hessians analysis on a multi-modal model trained on samples with complete multi-omics. The nodes represent biodomains, and the edges represent interactions, with the edge annotations being the rank of the interaction. The biodomain names based on the 3-letter abbreviations are: apoptosis (apo), lipid metabolism (lip), synapse (syn), endolysosome (end), metal binding (met), cell cycle (cel), mitochondrial metabolism (mit), myelination (mye), vasculature (vas), proteostasis (pro) and APP metabolism (app). The figure was rendered using functions from Matplotlib in Python.CC BY · PubMed Central
Fig. 7 Schematic of GNN-based feature extractor. The feature extractor is comprised of a stack of graph convolution blocks, a memory pooling layer, and a residual connection between the original input node features and the output of the memory pooling layer. Some elements of the figure were rendered using functions from Matplotlib in Python, while the entire figure was created in PowerPoint.
Fig. 7 Schematic of GNN-based feature extractor. The feature extractor is comprised of a stack of graph convolution blocks, a memory pooling layer, and a residual connection between the original input node features and the output of the memory pooling layer. Some elements of the figure were rendered using functions from Matplotlib in Python, while the entire figure was created in PowerPoint.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 69 references

Research works printed but not linked: 2

  1. Sundararajan, M., Taly, A. & Yan, Q. Axiomatic attribution for deep networks. in International Conference on Machine Learning, 2017: PMLR, 3319–3328.
  2. Janizek, J. D., Sturmfels, P. & Lee, S.-I. Explaining explanations: axiomatic feature interactions for deep networks. J. Mach. Learn. Res. 22, 1–54 (2021).

Cited by

See all 55

Metadata

Authors
Tripathy, R. K., Frohock, Z., Wang, H., Cary, G. A., Keegan, S., Carter, G. W., & Li, Y.
Affiliations
  • The Jackson Laboratory for Genomic Medicine (US)
  • Jackson Laboratory (US)
Corresponding author
  • Carter, Gregory W.
  • Li, Yi
Venue
npj Systems Biology and Applications
Published
2025-05-08
Volume
11
Issue
1
Pages
43-43
ISSN
2056-7189 (ISSN-L)
OpenAlex
W4410191441
Cited by
36
References
69
Citations by year
2026: 27 · 2025: 9
FWCI
10.82
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Systems Biology and Applications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • The Mouseion at the JAXlibrary (Jackson Laboratory) (submittedVersion, cc-by)
Funders
  • NIA (R21 AG083299)
  • NIH (U54 AG065187)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
  • Physical Sciences › Computer Science › Artificial Intelligence › Advanced Graph Neural Networks
Keywords
multi-omics integration, knowledge graphs, graph neural networks, explainable AI, biomarker discovery, Alzheimer's disease, transcriptomics, proteomics, drug target prediction
MeSH terms
Multiomics, Graph Neural Networks, Algorithms, Alzheimer Disease, Humans, Biomarkers, Neural Networks, Computer, Computational Biology, Gene Expression Profiling, Genomics, Proteomics, Transcriptome

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Effective integration of multi-omics with prior knowledge to identify biomarkers via explainable graph neural networks. https://app.cochat.ai/papers/10.1038/s41540-025-00519-9
The paper
Tripathy, R. K., Frohock, Z., Wang, H., Cary, G. A., Keegan, S., Carter, G. W., & Li, Y. (2025). Effective integration of multi-omics with prior knowledge to identify biomarkers via explainable graph neural networks. npj Systems Biology and Applications. https://doi.org/10.1038/s41540-025-00519-9

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.