CoChat
Sign Up
Cited by 294Advanced Graph Neural NetworksBioinformatics and Genomic NetworksComputational Drug Discovery Methods

Graph representation learning in biomedicine and healthcare

Open access
Li, Michelle M., Huang, Kexin, Zitnik, Marinka·2022·Nature Biomedical Engineering
DOI 10.1038/s41551-022-00942-xPMC PMC10699434
Open this paper in CoChat — free accountOpen PDF

Abstract

Networks—or graphs—are universal descriptors of systems of interacting elements. In biomedicine and healthcare, they can represent, for example, molecular interactions, signalling pathways, disease co-morbidities or healthcare systems. In this Perspective, we posit that representation learning can realize principles of network medicine, discuss successes and current limitations of the use of representation learning on graphs in biomedicine and healthcare, and outline algorithmic strategies that leverage the topology of graphs to embed them into compact vectorial spaces. We argue that graph representation learning will keep pushing forward machine learning for biomedicine and healthcare applications, including the identification of genetic variants underlying complex traits, the disentanglement of single-cell behaviours and their effects on health, the assistance of patients in diagnosis and treatment, and the development of safe and effective medicines. This Perspective outlines the successes and limitations of graph deep learning for biomedical and healthcare applications.

AI snapshot

Written from the paper’s full text.

TL;DR

This review argues that long-standing systems biology and medicine principles—such as the local hypothesis, shared components hypothesis, disease module hypothesis, and network parsimony—provide conceptual grounding for graph representation learning in biomedicine. It surveys shallow embeddings, graph neural networks, and generative graph models across molecular, genomic, therapeutic, and healthcare applications.

Key findings

  • Graph representation learning learns compact vector embeddings in which nodes with similar network neighborhoods are placed close together, and it can be viewed as a differentiable engine of network biomedicine principles including the local hypothesis, shared components hypothesis, disease module hypothesis, and network parsimony.
  • The reviewed methods span shallow graph embeddings (e.g., DeepWalk, node2vec, LINE, knowledge graph embeddings), graph neural networks with message passing, neighborhood aggregation, and representation updates, and deep generative graph models using variational autoencoders, generative adversarial networks, junction trees, and sequential node/edge sampling.
  • Biomedical applications covered include protein structure, protein interactions, and protein function; gene expression, single-cell RNA sequencing, and spatial molecular profiling; drug-target, drug-drug, and drug-disease prediction; and patient-specific networks from histopathology images and electronic health records.
  • No primary numerical results or benchmark performance numbers are reported in this review.

Methods

  • Design: narrative review/survey, not a primary empirical study; it synthesizes graph representation learning approaches and biomedical applications.
  • Data or sample: no new dataset is analyzed; the paper discusses published methods and application areas across molecular, genomic, therapeutic, and healthcare domains.
  • Main analysis: conceptual synthesis linking graph representation learning to systems biology and medicine principles, organized into shallow graph embedding approaches, graph neural networks, and generative graph models.

Limitations

  • Stated in the paper: graphs contain complex topographical structure, have no fixed node ordering or reference points, and comprise many different entity and edge types, which classic deep learning methods are unable to handle directly.
  • Stated in the paper: biomedical networks are often incomplete or limited—for example, PPI networks are limited in nodes and edges due to experimental and resource constraints, gene expression data can be noisy and variational, and PPI networks cannot capture all gene regulatory activities.
  • Evident from the text: as a review, it provides no empirical benchmark, systematic search protocol, or quantitative comparison; it also notes that explainability, fairness, and robustness of graph representations remain essential but unresolved needs in clinical applications.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41551-022-00942-x

Cites

199 of 199 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 169 cited sources checked · 0 supported · 0 flagged

  • Premisereasoning

    Unlike shallow embeddings, GNNs can capture higher-order and non-linear patterns through multi-hop propagation across several layers of neural message passing.

    “In contrast to shallow embeddings, GNNs can capture higher-order and non-linear patterns through multi-hop propagation within several layers of neural message passing.”
    Quote in the paperNot a substantial claim
  • Premisecitation

    Under the shared components and disease module hypotheses, cellular components linked to the same phenotype tend to cluster in the same network neighborhood.

    “Given by the shared components and disease module hypotheses [13], cellular components associated with the same phenotype tend to cluster in the same network neighborhood [16].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [13] Barabási, A.-L., Gulbahce, N. & Loscalzo, J. Network medicine: a network-based approach to human disease. Nat. Rev. Genet. 12, 56–68 (2011). source

      Not checked in the free run.

    • cites [16] Agrawal, M., Zitnik, M. & Leskovec, J. Large-scale analysis of disease pathways in the human interactome. In Pac. Symp. Biocomput. 23, 111–122 (2018).

      Not checked in the free run.

  • Premisecitation

    Topology-based methods have been shown to capture and use biological system dynamics to enrich existing PPI networks.

    “Topology-based methods have been shown to capture and leverage the dynamics of biological systems to enrich existing PPI networks [119].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [119] Moreau, Y. & Tranchevent, L.-C. Computational tools for prioritizing candidate genes: boosting disease gene discovery. Nat. Rev. Genet. 13, 523–536 (2012). source

      Not checked in the free run.

  • Premisecitation

    The inductive ability of graph convolution neural networks to generalize to unseen data and to generate new data by decoding latent representations has enabled discovery of new molecules, interactions, and functions.

    “Specifically, the inductive ability of graph convolution neural networks to generalize to data points unseen during model training (Section 2.2) and even generate new data points from scratch by decoding latent representation from the embedding space (Section 2.3) has enabled the discovery of new molecules, interactions, and functions [101, 108, 109].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [101] Huang, K., Xiao, C., Glass, L. M., Zitnik, M. & Sun, J. SkipGNN: predicting molecular interactions with skip-graph networks. Sci. Rep. 10, 21092 (2020). source

      Not checked in the free run.

    • cites [108] Fout, A., Byrd, J., Shariat, B. & Ben-Hur, A. Protein interface prediction using graph convolutional networks. In Proc. International Conference on Neural Information Processing Systems (eds Guyon, I. et al.) (Curran Associates, 2019).

      Not checked in the free run.

    • cites [109] Ingraham, J., Garg, V. K., Barzilay, R. & Jaakkola, T. S. Generative models for graph-based protein design. In Proc. International Conference on Neural Information Processing Systems (eds Wallach, H. et al.) (Curran Associates, 2019).

      Not checked in the free run.

  • Interpretationnone given

    Node embeddings can capture whether a node lies in a hub based on its degree.

    “node embeddings can capture whether the nodes lie within a hub based on their degree, an important aspect of local neighborhood (e.g., strongly clustered gene embeddings indicate essential housekeeping roles)”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim. 42
  • May combine two claims. 27

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisereasoning

    Diagnosing diseases therefore requires knowledge of changes in coding gene transcription to capture genome-wide associations that drive disease acquisition and progression.

    “As a result, diagnosing diseases requires knowledge about alterations in the transcription of coding genes to capture genome-wide associations driving disease acquisition and progression.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisenone given

    Graph representation learning is difficult because graphs have complex topographical structure, lack fixed node ordering or reference points, and contain many entity and interaction types.

    “Graph representation learning is challenging because graphs contain complex topographical structure, have no fixed node ordering or reference points, and are comprised of many different kinds of entities (nodes) and various types of interactions (edges) relating them to each other.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    Representation learning has markedly advanced the state of the art in biomedical machine learning.

    “We also capture the breadth of ways in which representation learning has dramatically improved the state-of-the-art in biomedical machine learning.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Premisenone given

    Biomedical networks or graphs are universal descriptors for systems of interacting elements across molecular interactions, disease co-morbidity, healthcare systems, and scientific knowledge.

    “Biomedical networks (or graphs) are universal descriptors for systems of interacting elements, from molecular interactions and disease co-morbidity to healthcare systems and scientific knowledge.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    Shallow embedding approaches are transductive, with an encoder function that is a simple embedding lookup.

    “These approaches are transductive embedding methods where the encoder function is a simple embedding lookup (Figure 2).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    Graph neural networks are a class of neural networks designed for graph-structured datasets.

    “Graph neural networks (GNNs) are a class of neural networks designed for graph-structured datasets (Figure 2).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Essential genes are frequently located in molecular network hubs, while non-essential genes, including disease-associated ones, tend to lie on the periphery.

    “Further, essential genes are often found in hubs of a molecular network whereas non-essential genes (e.g., those associated with disease) are located on the periphery [13].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [13] Barabási, A.-L., Gulbahce, N. & Loscalzo, J. Network medicine: a network-based approach to human disease. Nat. Rev. Genet. 12, 56–68 (2011). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given

    These functions are optimized so that nodes with similar network neighborhoods are embedded near each other, and algebraic operations in the learned space reflect the network's topology.

    “Such functions are optimized to embed the input network so that nodes with similar network neighborhoods are embedded close together in the embedding space, and algebraic operations performed in this learned space reflect the network’s topology.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationnone given

    The paper's central observation is that long-standing systems biology and medicine principles, though often unstated in machine learning, ground graph representation learning, account for its current successes and limitations, and can guide future advances.

    “we put forward an observation that long-standing principles of systems biology and medicine—while often unspoken in machine learning research—provide the conceptual grounding for representation learning on graphs, explain its current successes and limitations, and even inform future advancements.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisereasoning

    Graph representation learning methods enable analysis of heterogeneous multimodal networks and predictions across domains from genomic data to clinically relevant data.

    “Graph representation learning methods allow us to analyze heterogeneous networks of multimodal data and make predictions across domains, from genomic level data (e.g., gene expression, copy number information) to clinically relevant data (e.g., pathophysiology, tissue localization).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    The core idea of representation learning is to represent nodes or larger graph structures as points in a low-dimensional space whose geometry is optimized to reflect interaction structure.

    “Its core idea is to learn how to represent nodes (or larger graph structures) in a network as points in a low-dimensional space, where the geometry of this space is optimized to reflect the structure of interactions between nodes.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    Comparing transcriptomic profiles of healthy individuals with those of patients who have a specific disease informs clinicians about its causal genes.

    “Comparing transcriptomic profiles from healthy individuals to those of patients with a specific disease informs clinicians of its causal genes.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisenone given

    Graph representation learning approaches fundamentally use graph topology to embed networks into compact vector spaces.

    “We synthesize a spectrum of algorithmic approaches that, at their core, leverage graph topology to embed networks into compact vector spaces.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    GNNs are commonly used to generate protein representations from chemical and geometric features to predict protein pocket-ligand and protein-protein interaction sites, intra- and inter-molecular energies, binding affinities, complex quality measures, and interface residues.

    “GNNs (as described in Section 4.1) are commonly used to generate representations of proteins based on chemical (e.g., free electrons’ and protons donors’ locations) and geometric (e.g., distance-dependent curvature) features to predict protein pocket-ligand and protein-protein interactions sites [116]; intra- and inter-molecule residue contact graphs to predict intra- and inter-molecular energies,”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [116] Li, D. & Gao, J. Towards perturbation prediction of biological networks using deep learning. Sci. Rep. 9, 11941 (2019). source

      Not checked in the free run.

    • cites [117] Liu, Y., Yuan, H., Cai, L., & Ji, S. Deep learning of high-order interactions for protein interface prediction. In Proc. ACM SIGKDD Conference on Knowledge Discovery and Data Mining 679–687 (ACM, 2020). source

      Not checked in the free run.

    • cites [111] Elton, D. C., Boukouvalas, Z., Fuge, M. D. & Chung, P. W. Deep learning for molecular design—a review of the state of the art. Mol. Syst. Des. Eng. 4, 828–849 (2019). source

      Not checked in the free run.

  • May combine two claims.

    Premisenone given

    Adding gene expression data to PPI networks has identified disease biomarkers, which are then used to classify diseases of interest more accurately.

    “Further, injecting gene expression data into PPI networks has identified disease biomarkers, which are then used to more accurately classify diseases of interest.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    GNNs learn compact representations of graph elements, their attributes, and supervised labels when present.

    “They learn compact representations of graph elements, their attributes, and supervised labels if any.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationreasoning

    Graph representation learning is positioned to transform the study of complex systems, analogous to how deep learning transformed image analysis and natural language processing.

    “Akin to how deep learning on images and sequences has revolutionized image analysis and natural language processing, graph representation learning is poised to transform the study of complex systems.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    Shallow embedding methods optimize a compact vector space so that points close in the graph map to nearby points in the embedding space, measured by a predefined distance function or an outer product.

    “Shallow embedding methods optimize a compact vector space such that points that are close in the graph are mapped to nearby points in the embedding space, which is measured by a predefined distance function or an outer product.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisereasoning

    Deep generative models estimate distributional graph properties from a graph dataset and infer graph structures using those optimized distributions.

    “Deep generative models address the challenge by estimating distributional graph properties based on a dataset of graphs 𝒢 and inferring graph structures using such optimized distributions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    GNNs can optimize supervised signals and graph structure at the same time, whereas shallow embedding methods need a two-stage approach for the same.

    “Additionally, GNNs can optimize supervised signals and graph structure simultaneously, whereas shallow embedding approaches require a two-stage approach to achieve the same.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisereasoning

    Because protein structures fold into complex 3D shapes, representing them as graphs is natural.

    “Since protein structures are folded into complex 3D structures, it is natural to represent them as graphs.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Gene Ontology terms are a standardized vocabulary for molecular functions, biological processes, and cellular locations of gene products.

    “Gene Ontology (GO) terms [125] are a standardized vocabulary for describing molecular functions, biological processes, and cellular locations of gene products [126].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [125] Hasibi, R. & Michoel, T. A Graph feature auto-encoder for the prediction of unobserved node features on biological networks. BMC Bioinform 22, 525 (2021). source

      Not checked in the free run.

    • cites [126] Cao, M. et al. Going the distance for protein function prediction: a new distance metric for protein interaction networks. PLoS ONE 8, e76339 (2013). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Learning the topology of graphs with drug, protein, and disease nodes has improved prediction of candidate drugs, identification of off-target effects, and prioritization of novel drug combinations.

    “Learning the topology of graphs containing drug, protein, and disease nodes has improved predictions of candidate drugs for treating a disease, identification of potential off-target effects, and prioritization of novel drug combinations.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [103] Fan, K. & Zhang, Y. Pseudo2GO: A graph-based deep learning method for pseudogene function prediction by borrowing information from coding genes. Front. Genet. 11, 807 (2020). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    AI methods that produce representations can be viewed as differentiable engines of key network biomedicine principles.

    “Hence, artificial intelligence methods that produce representations can be thought of as differentiable engines of key network biomedicine principles.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    Diseases are classified by patients' presenting symptoms, which can result from molecular dysfunctions such as genetic mutations.

    “Diseases are classified based on the presenting symptoms of patients, which can be caused by molecular dysfunctions, such as genetic mutations.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisenone given

    Representation learning uses deep, non-linear transformation functions that map nodes to points in a compact vector space called embeddings.

    “More concretely, representation learning specifies deep, non-linear transformation functions that map nodes to points in a compact vector space, termed embeddings.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    Because biomedical networks can be multimodal and massive, scaling GNNs to large and heterogeneous networks requires special consideration.

    “As biomedical networks can be multimodal and massive, special consideration is needed to scale GNNs to large and heterogeneous networks.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    GO terms have been built as a hierarchical graph that GNNs use to learn term dependencies, or they are used directly as protein function labels.

    “They have been built as a hierarchical graph that GNNs then leverage to learn dependencies of the terms [126], or directly used as protein function labels [106, 127].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [126] Cao, M. et al. Going the distance for protein function prediction: a new distance metric for protein interaction networks. PLoS ONE 8, e76339 (2013). source

      Not checked in the free run.

    • cites [106] Hamilton, W. L., Ying, Z. & Leskovec, J. Inductive representation learning on large graphs. In Proc. International Conference on Neural Information Processing Systems (eds Guyon, I. et al.) (Curran Associates, 2017).

      Not checked in the free run.

    • cites [127] Dey, T. K. & Mandal, S. Protein classification with improved topological data analysis. In 18th International Workshop on Algorithms in Bioinformatics (WABI) (eds Parida, L. & Ukkonen, E.) 6:1–6:13 (Schloss Dagstuhl–Leibniz-Zentrum fuer Informatik, 2018).

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Small compounds are modeled as 2D and 3D molecular graphs with atoms as nodes and bonds as edges.

    “Similar to proteins, small compounds are modeled as 2D and 3D molecular graphs such that nodes are atoms and edges are bonds.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [83] Schlichtkrull, M. et al. Modeling relational data with graph convolutional networks. In European Semantic Web Conference (eds Gangemi, A. et al.) 593–607 (Springer, 2018). source

      Not checked in the free run.

    • cites [166] Wang, R., Li, S., Cheng, L., Wong, M. H. & Leung, K. S. Predicting associations among drugs, targets and diseases by tensor decomposition for drug repositioning. BMC Bioinf. 20, 628 (2019). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Graph representation learning applied to gene or cell similarity networks can impute single-cell RNA sequencing data and predict cell clusters.

    “Applied to such networks, graph representation learning can impute scRNA-seq data [156, 157], predict cell clusters [157, 158], etc.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [156] Alagappan, M., Jiang, D., Denko, N. & Koong, A. C. in Tumor Microenvironment (eds Koumenis, C. et al.) 253–268 (Springer, 2016). source

      Not checked in the free run.

    • cites [157] Thafar, M. A. et al. DTiGEMS+: drug–target interaction prediction using graph embedding, graph mining, and similarity-based techniques. J. Cheminformatics 12, 44 (2020). source

      Not checked in the free run.

    • cites [158] Thafar, M. A. et al. DTi2Vec: Drug–target interaction prediction using network embedding and ensemble learning. J. Cheminformatics 13, 71 (2021). source

      Not checked in the free run.

    • Rests on an assumption: A predominant approach to analyze single-cell RNA sequencing datasets is to transform them into gene similarity networks or cell similarity networks by correlating gene expression readouts across individual cells.

  • May combine two claims.

    Premisecitation

    Graph representation learning methods such as GNNs have been developed to learn robust and meaningful molecular representations despite the incomplete interactome and to inductively infer new edges between node pairs.

    “To this end, graph representation learning methods, such as GNNs, have been developed to learn robust and meaningful representations of molecules despite the incomplete interactome [102] and to inductively infer new edges between pairs of nodes [154].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [102] Yin, N. et al. Synergistic and antagonistic drug combinations depend on network topology. PLoS ONE 9, e93960 (2014). source

      Not checked in the free run.

    • cites [154] Coley, C. W. et al. A graph-convolutional neural network model for the prediction of chemical reactivity. Chem. Sci. 10, 370–377 (2019). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Spatial molecular profiling can measure gene expression at the cellular level and the location of cells in a tissue.

    “Spatial molecular profiling can measure both gene expression at the cellular level and location of cells in a tissue [160].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [160] Jiang, M. et al. Drug–target affinity prediction using graph neural network and contact maps. RSC Adv 10, 20701–20712 (2020). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Edges in molecular graphs can be added to indicate pairwise spatial distance between atoms, or directed with bond angles and rotations incorporated.

    “Edges can also be added to indicate pairwise spatial distance between two atoms [72], or directed with information on bond angles and rotations incorporated into the molecular graph [84].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [72] Hu, Z., Dong, Y., Wang, K. & Sun, Y. Heterogeneous graph transformer. In Proc. ACM Web Conference 2704–2710 (ACM, 2020). source

      Not checked in the free run.

    • cites [84] Wang, X. et al. Heterogeneous graph attention network. In Proc. ACM Web Conference (eds Liu, L. & White, R.) 2022–2032 (ACM, 2019). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Gene co-expression and molecular similarity graphs require a threshold applied to gene expression data.

    “Creating cell neighborhood and spatial gene expression graphs require a distance metric, as edges are determined based on spatial proximity, while gene co-expression and molecular similarity graphs need a threshold applied on the gene expression data [35].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [35] Hetzel, L., Fischer, D. S., Günnemann, S. & Theis, F. J. Graph representation learning for single cell biology. Curr. Opin. Syst. Biol. 28, 100347 (2021). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Spatial transcriptomics data can be used to construct cell graphs, spatial gene expression graphs, gene co-expression networks, or molecular similarity graphs.

    “As a result, spatial transcriptomics data can be used to construct cell graphs [161], spatial gene expression graphs [162], gene co-expression networks, or molecular similarity graphs [35].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [161] Chen, L. et al. TransformerCPI: improving compound–protein interaction prediction by sequence-based deep learning with self-attention mechanism and label reversal experiments. Bioinformatics 36, 4406–4414 (2020). source

      Not checked in the free run.

    • cites [162] Quan, Z., Guo, Y., Lin, X., Wang, Z.-J. & Zeng, X. GraphCPI: Graph neural representation learning for compound-protein interaction. In BIBM 717–722 (2019). source

      Not checked in the free run.

    • cites [35] Hetzel, L., Fischer, D. S., Günnemann, S. & Theis, F. J. Graph representation learning for single cell biology. Curr. Opin. Syst. Biol. 28, 100347 (2021). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Modern drug discovery requires elucidating a candidate drug's chemical structure, identifying its targets, quantifying efficacy and toxicity, and detecting side effects.

    “Modern drug discovery requires elucidating a candidate drug’s chemical structure, identifying its drug targets, quantifying its efficacy and toxicity, and detecting its potential side effects [13, 14, 32, 164].”
    Quote in the paperSubstantial claim0 of 4 sources checked
    • cites [13] Barabási, A.-L., Gulbahce, N. & Loscalzo, J. Network medicine: a network-based approach to human disease. Nat. Rev. Genet. 12, 56–68 (2011). source

      Not checked in the free run.

    • cites [14] Hu, J. X., Thomas, C. E. & Brunak, S. Network biology concepts in complex disease comorbidities. Nat. Rev. Genet. 17, 615–629 (2016). source

      Not checked in the free run.

    • cites [32] Rai, A., Shinde, P. & Jalan, S. Network spectra for drug-target identification in complex diseases: new guns against old foes. Appl. Netw. Sci. 3, 51 (2018). source

      Not checked in the free run.

    • cites [164] Lin, X. et al. DeepGS: Deep representation learning of graphs and sequences for drug-target binding affinity prediction. ECAI (2020).

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Each atom and bond in a molecular graph may include features such as atomic mass, atomic number, and bond type.

    “Each atom and bond may include features, such as atomic mass, atomic number, and bond type, to be included in the model [83, 166].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [83] Schlichtkrull, M. et al. Modeling relational data with graph convolutional networks. In European Semantic Web Conference (eds Gangemi, A. et al.) 593–607 (Springer, 2018). source

      Not checked in the free run.

    • cites [166] Wang, R., Li, S., Cheng, L., Wong, M. H. & Leung, K. S. Predicting associations among drugs, targets and diseases by tensor decomposition for drug repositioning. BMC Bioinf. 20, 628 (2019). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Two separate GNNs may be used to learn representations of protein and small molecule graphs for predicting drug-target affinity.

    “Also, two separate GNNs may be used to learn representations of protein and small molecule graphs for predicting drug-target affinity [173].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [173] Choi, E., Bahadori, M. T., Song, L., Stewart, W. F. & Sun, J. GRAM: graph-based attention model for healthcare representation learning. In Proc. 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 787–795 (ACM, 2017). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Medical images of patients, including histopathology slides, enable clinicians to comprehensively observe the effects of a disease on the patient's body.

    “Medical images of patients, including histopathology slides, enable clinicians to comprehensively observe the effects of a disease on the patient’s body [188].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [188] Wu, T. et al. Representation learning of EHR data via graph-based medical entity embedding. In Advances in Neural Information Processing Systems Graph Representation Learning Workshop (2019).

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    MeSH terms may be used to construct a drug-disease graph, from which latent representations of drugs and diseases are learned using graph embedding algorithms including DeepWalk and LINE.

    “For instance, Medical Subject Headings (MeSH) terms may be used to construct a drug-disease graph, from which latent representations of drugs and diseases are learned using various graph embedding algorithms, including DeepWalk and LINE (Section 2.1) [178].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [178] Zhou, Y. et al. CGC-Net: cell graph convolutional network for grading of colorectal cancer histology images. In IEEE International Conference on Computer Vision Workshops (IEEE, 2019). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    TDA transforms experimental data into a graph where nodes are compounds and edges indicate similarity.

    “Concretely, TDA (refer to SI Note 3) transforms experimental data into a graph where nodes represent compounds and edges indicate a level of similarity between them [169].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [169] Alsentzer, E., Finlayson, S. G., Li, M. M. & Zitnik, M. Subgraph neural networks. In Proc. International Conference on Neural Information Processing Systems (eds Larochelle, H. et al.) 8017–8029 (Curran Associates, 2020).

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Graph representation learning enables in silico modeling of drug action, allowing more efficient ranking of candidate drugs for repurposing by considering gene expression, gene ontologies, drug similarity, and clinical data.

    “Graph representation learning methods enables in silico modeling of drug action, which allows for more efficient ranking of candidate drugs for repurposing, such as by considering gene expression data, gene ontologies, drug similarity, and other clinically relevant data regarding side effects and indications.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    K-nearest neighbors is a common method to construct drug and target similarity networks.

    “K-nearest neighbors, for instance, is a common used method to construct drug and target similarity networks [171].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [171] Barisoni, L., Lafata, K. J., Hewitt, S. M., Madabhushi, A. & Balis, U. G. Digital pathology and computational image analysis in nephropathology. Nat. Rev. Nephrol. 16, 669–685 (2020). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    A cell-to-tissue graph can capture cell morphology and interactions, tissue morphology and spatial distribution, cell-to-tissue hierarchies, and spatial distribution of cells with respect to tissues, and a hierarchical GNN can learn representations from these modalities.

    “Further, cell morphology and interactions, tissue morphology and spatial distribution, cell-to-tissue hierarchies, and spatial distribution of cells with respect to tissues can be captured in a cell-to-tissue graph, upon which a hierarchical GNN can learn representations using these different data modalities [189].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [189] Mao, C., Yao, L. & Luo, Y. MedGCN: medication recommendation and lab test imputation via graph convolutional networks. J. Biomed. Inform. 127, 104000 (2022). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Message passing steps in GNNs aggregate information from neighboring atoms and bonds to learn the local chemistry of each atom.

    “Intuitively, message passing steps (i.e., in GNNs (Section 2.2)) aggregate information from neighboring atoms and bonds to learn the local chemistry of each atom [166].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [166] Wang, R., Li, S., Cheng, L., Wong, M. H. & Leung, K. S. Predicting associations among drugs, targets and diseases by tensor decomposition for drug repositioning. BMC Bioinf. 20, 628 (2019). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Minimizing adverse drug events is part of the drug discovery pipeline.

    “Part of the drug discovery pipeline is minimizing adverse drug events [33, 34].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [33] David, L., Thakkar, A., Mercado, R. & Engkvist, O. Molecular representations in AI-driven drug discovery: a review and practical guide. J. Cheminformatics 12, 56 (2020). source

      Not checked in the free run.

    • cites [34] Wieder, O. et al. A compact review of molecular property prediction with graph neural networks. Drug Discov. Today. Technol. 37, 1–12 (2020). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    EHRs are often supplemented with other modalities such as diseases, symptoms, molecular data, drug interactions, etc.

    “EHRs are often supplemented with other modalities, such as diseases, symptoms, molecular data, drug interactions, etc [7, 206, 211, 212].”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [7] Nelson, C. A., Butte, A. J. & Baranzini, S. E. Integrating biomedical research and electronic health records to create knowledge-based biologically meaningful machine-readable embeddings. Nat. Commun. 10, 3045 (2019). source

      Not checked in the free run.

    • cites [206] Lanchantin, J. & Qi, Y. Graph convolutional networks for epigenetic state prediction using both sequence and 3D genome data. Bioinformatics 36, i659–i667 (2020). source

      Not checked in the free run.

    • cites [211] Ying, Z., Bourgeois, D., You, J., Zitnik, M. & Leskovec, J. GNNExplainer: generating explanations for graph neural networks. In Proc. International Conference on Neural Information Processing Systems (ed Wallach, H. et al.) (Curran Associates, 2019).

      Not checked in the free run.

    • cites [212] Agarwal, C., Lakkaraju, H. & Zitnik, M. Towards a unified framework for fair and stable graph representation learning. In Proc. Machine Learning Research (eds de Campos, C. & Maathuis, M. H.) 2114–2124 (PMLR, 2021).

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    A patient graph may be constructed based on patient similarity, and patient embeddings learned by an LSTM-GNN architecture can be optimized to predict patient outcomes.

    “Alternatively, a patient graph may be constructed based on the similarity of patients, and patient embeddings learned by an LSTM-GNN architecture are optimized to predict patient outcomes [210].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [210] Machens, A. et al. An infectious disease model on empirical networks of human contact: bridging the gap between dynamic network data and contact matrices. BMC Infect. Dis. 13, 185 (2013). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    An ST-GCN is designed to utilize underlying spatial and temporal dependencies of EHR data for generating patient diagnoses.

    “An ST-GCN [78] is designed to utilize the underlying spatial and temporal dependencies of EHR data for generating patient diagnoses [187].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [187] Rotmensch, M., Halpern, Y., Tlimat, A., Horng, S. & Sontag, D. Learning a health knowledge graph from electronic medical records. Sci. Rep. 7, 5994 (2017). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    TDA can generate graphs of whole-slide images including tissues from various patient sources, and GNNs can classify the stage of colon cancer.

    “Alternatively, TDA can generate graphs of whole-slide images, which include tissues from various patient sources (SI Note 3), and GNNs are then used to classify the stage of colon cancer [199].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [199] Li, L. et al. A method to learn embedding of a probabilistic medical knowledge graph: algorithm development. JMIR Med. Inf. 8, e17645 (2020). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Meta-paths may be exploited in an EHR-derived KG to leverage higher order, semantically important relations for disease classification.

    “Meta-paths may alternatively be exploited in an EHR-derived KG to leverage higher order, semantically important relations for disease classification [214].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [214] Obermeyer, Z., Powers, B., Vogeli, C. & Mullainathan, S. Dissecting racial bias in an algorithm used to manage the health of populations. Science 366, 447–453 (2019). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Node features for drugs and diseases can be initialized using Skipgram and then a GNN with multi-layer message-passing can predict adverse drug events.

    “Node features for drugs and diseases can be initialized using Skipgram and then a GNN leveraging multi-layer message-passing can be applied to predict adverse drug events [211].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [211] Ying, Z., Bourgeois, D., You, J., Zitnik, M. & Leskovec, J. GNNExplainer: generating explanations for graph neural networks. In Proc. International Conference on Neural Information Processing Systems (ed Wallach, H. et al.) (Curran Associates, 2019).

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    In public health, spatial and temporal networks can model space- and time-dependent observations to spot trends, detect anomalies, and interpret temporal dynamics.

    “In public health, spatial and temporal networks can model space- and time-dependent observations (e.g., disease states, susceptibility to infection [225]) to spot trends, detect anomalies, and interpret temporal dynamics.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [225] Tran, D. et al. Fast and precise single-cell data analysis using a hierarchical autoencoder. Nat. Commun. 12, 1029 (2021). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    GNNs can generate cell spatial graphs from histopathology images and fuse genomic and transcriptomic data to predict treatment response and resistance, histopathology grading, and patient survival.

    “Since multimodal data enables more robust predictions, GNNs (Section 2.2) have been applied to generate cell spatial graphs from histopathology images and then fuse genomic and transcriptomic data for predicting treatment response and resistance, histopathology grading, and patient survival [8].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [8] Chen, R. J. et al. Pathomic fusion: an integrated framework for fusing histopathology and genomic features for cancer diagnosis and prognosis. In IEEE Transactions on Medical Imaging Vol. 41, 757–770 (IEEE, 2022). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Attention mechanisms may be applied on EHR data and medical ontologies to capture parent-child relationships.

    “Alternatively, attention mechanisms may be applied on EHR data and medical ontologies to capture the parent-child relationships [186, 204, 205].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [186] Hu, J. et al. SpaGCN: integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network. Nat. Methods 18, 1342–1351 (2021). source

      Not checked in the free run.

    • cites [204] Wang, T., Peng, Q., Liu, B., Liu, Y. & Wang, Y. Disease module identification based on representation learning of complex networks integrated from GWAS, eQTL summaries, and human interactome. Front. Bioeng. Biotechnol. 8, 418 (2020). source

      Not checked in the free run.

    • cites [205] Dekker, J. & Misteli, T. Long-range chromatin interactions. Cold Spring Harb. Perspect. Biol. 7, a019356 (2015). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationcitation

    Graphs can model long-range dependencies or interactions, so chromatin elements and effects of their binding to regions across the genome can be modeled as a network.

    “Additionally, because graphs can model long-range dependencies or interactions, we can model chromatin elements and the effects of their binding to regions across the genome as a network [220, 221].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [220] Stark, C. et al. BioGRID: a general repository for interaction datasets. Nucleic Acids Res 34, D535–D539 (2006). source

      Not checked in the free run.

    • cites [221] Efremova, M., Vento-Tormo, M., Teichmann, S. A. & Vento-Tormo, R. Cellphonedb: inferring cell–cell communication from combined expression of multi-subunit ligand–receptor complexes. Nat. Protoc. 15, 1484–1506 (2020). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A probabilistic KG of EHR data including medical history, drug prescriptions, and laboratory examination results has been used to consider semantic relations between EHR entities in a shallow network embedding method.

    “A probabilistic KG of EHR data, which include medical history, drug prescriptions, and laboratory examination results, has been used to consider the semantic relations between EHR entities in a shallow network embedding method (Section 2.1) [213].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [213] Zhang, X. & Zitnik, M. GNNGuard: defending graph neural networks against adversarial attacks. In Proc. International Conference on Neural Information Processing Systems (eds Larochelle, H. et al.) 9263–9275 (Curran Associates, 2020).

      Not checked in the free run.

57 of 57

  • The paper relies on this without support.

    A predominant approach to analyze single-cell RNA sequencing datasets is to transform them into gene similarity networks or cell similarity networks by correlating gene expression readouts across individual cells.

    “In particular, a predominant approach to analyze scRNA-seq datasets is to transform them into gene similarity networks, such as gene co-expression networks, or cell similarity networks by correlating gene expression readouts across individual cells.”

  • The paper relies on this without support.

    Cross-domain expertise is necessary to develop a drug with optimal binding affinity and specificity to biomarkers, maximal therapy efficacy, and minimal adverse effects.

    “However, cross-domain expertise is necessary to develop a drug with the optimal binding affinity and specificity to biomarkers, maximal therapy efficacy, and minimal adverse effects.”

  • The paper relies on this without support.

    It is critical to integrate chemical structure information, protein interactions, and clinically relevant data into predictive models for drug discovery and repurposing.

    “As a result, it is critical to integrate chemical structure information, protein interactions, and clinically relevant data (e.g., indications and reported side effects) into predictive models for drug discovery and repurposing.”

  • The paper relies on this without support.

    The hierarchical information inherent to ICD codes naturally lends itself to creating a rich network of medical knowledge.

    “The hierarchical information inherent to ICD codes (medical ontologies) naturally lend itself to creating a rich network of medical knowledge.”
4 of 4

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?

  • In biomedical applications, graph structures can represent various associations, including gene functions, drug responses, medical image patches, and cellular similarity.

    “In the context of biomedical applications, these graph structures can represent various associations, including gene functions, drug responses, medical image patches, and cellular similarity [64].”

    From Challenges in AI-driven Biomedical Multimodal Data Fusion and Analysis (2025) · cites this paper as [64]

  • Geometric deep-learning models optimized on large medical KGs can match disease signatures to therapeutic candidates based on networks perturbed in disease.

    “To provide therapeutic predictions, geometric deep-learning models optimized on large medical KGs15 can match disease signatures to therapeutic candidates based on networks perturbed in disease15–18.”

    From A foundation model for clinician-centered drug repurposing (2024) · cites this paper as [15]

  • TxGNN achieves state-of-the-art performance in drug repurposing by leveraging a network medicine principle focusing on disease-treatment mechanisms.

    “TxGNN achieves state-of-the-art performance in drug repurposing by leveraging a network medicine principle focusing on disease–treatment mechanisms15.”

    From A foundation model for clinician-centered drug repurposing (2024) · cites this paper as [15]

  • The standard benchmarking strategy used to evaluate drug-repurposing AI models aligns with the approach predominantly used in the literature.

    “This evaluation method aligns with the approach predominantly used in the literature12,14–16,18–20.”

    From A foundation model for clinician-centered drug repurposing (2024) · cites this paper as [15]

Self-check

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

Each contradiction between two passages listed here was confirmed by an independent check.

  • A cell similarity graph is described as connecting genes

    Minor

    The text defines cell similarity networks in terms of relationships among cells, then describes creating such a graph by connecting genes rather than cells.

1 of 1

Cites

Flagged first, then by citations

19 of the 218 references couldn’t be matched to a known paper.

Research works printed but not linked: 39

  1. Hamilton, W. L., Ying, R. & Leskovec, J. Representation learning on graphs: methods and applications. IEEE Data Eng. Bull 40, 52–74 (2017).
  2. Hamilton, W. L. in Synthesis Lectures on Artificial Intelligence and Machine Learning Vol. 14, 1–159 (Morgan and Claypool, 2020).
  3. Kazemi, S. M. et al. Representation learning for dynamic graphs: a survey. J Mach. Learn. Res. 21, 1–73 (2020).
  4. Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S. & Dean, J. Distributed representations of words and phrases and their compositionality. In Proc. 26th International Conference on Neural Information Processing Systems Vol. 2 (eds Burges, C. J. et al.) 3111–3119 (Curran Associates, 2013).
  5. Bordes, A., Usunier, N., García-Durán, A., Weston, J. & Yakhnenko, O. Translating embeddings for modeling multi-relational data. In Proc. 26th International Conference on Neural Information Processing Systems Vol. 2 (eds Burges, C. J. et al.) 2787–2795 (Curran Associates, 2013).
  6. Nickel, M., Tresp, V. & Kriegel, H. A three-way model for collective learning on multi-relational data. In Proc. 28th International Conference on International Conference on Machine Learning 809–816 (PMLR, 2011).
  7. Trouillon, T., Welbl, J., Riedel, S., Gaussier, É. & Bouchard, G. Complex embeddings for simple link prediction. In Proc. 33rd International Conference on International Conference on Machine Learning Vol. 48 (eds Balcan, M. F. & Weinberger, K. Q.) 2071–2080 (PMLR, 2016).
  8. Sun, Z., Deng, Z., Nie, J. & Tang, J. RotatE: knowledge graph embedding by relational rotation in complex space. In International Conference on Learning Representations (2019).
  9. Yang, B., Yih, W., He, X., Gao, J. & Deng, L. Embedding entities and relations for learning and inference in knowledge bases. In International Conference on Learning Representations (2015).
  10. Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O. & Dahl, G. E. Neural message passing for quantum chemistry. In Proc. 34th International Conference on Machine Learning Vol. 70, 1263–1272 (PMLR, 2017).
  11. Kipf, T. N. & Welling, M. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations (2017).
  12. Xu, K., Hu, W., Leskovec, J. & Jegelka, S. How powerful are graph neural networks? In International Conference on Learning Representations (2019).
  13. Vinyals, O., Bengio, S. & Kudlur, M. Order matters: sequence to sequence for sets. In International Conference on Learning Representations (2016).
  14. Defferrard, M., Bresson, X. & Vandergheynst, P. Convolutional neural networks on graphs with fast localized spectral filtering. In Proc. International Conference on Neural Information Processing Systems (eds Lee, D. et al.) 3844–3852 (Curran Associates, 2016).
  15. Yun, S., Jeong, M., Kim, R., Kang, J. & Kim, H. J. Graph transformer networks. In Proc. 33rd Conference on Neural Information Processing Systems (eds Wallach, H. et al.) (Curran Associates, 2019).
  16. Klicpera, J., Groß, J. & Günnemann, S. Directional message passing for molecular graphs. In International Conference on Learning Representations (2020).
  17. Zeng, H., Zhou, H., Srivastava, A., Kannan, R. & Prasanna, V. K. GraphSAINT: graph sampling based inductive learning method. In International Conference on Learning Representations (2020).
  18. Huang, K. & Zitnik, M. Graph meta learning via local subgraphs. In Proc. International Conference on Neural Information Processing Systems (eds Larochelle, H. et al.) 5862–5874 (Curran Associates, 2020).
  19. Hu, W. et al. Strategies for pre-training graph neural networks. In International Conference on Learning Representations (2020).
  20. Barabási A.-L. et al. Network Science (Cambridge Univ. Press, 2016).
  21. Jin, W., Barzilay, R. & Jaakkola, T. S. Junction tree variational autoencoder for molecular graph generation. In International Conference on Machine Learning (PMLR, 2018).
  22. Kipf, T. N. & Welling, M. Variational graph auto-encoders. In Advances in Neural Information Processing Systems Bayesian Deep Learning Workshop (2016).
  23. You, J., Liu, B., Ying, R., Pande, V. & Leskovec, J. Graph convolutional policy network for goal-directed molecular graph generation. In Proc. International Conference on Neural Information Processing Systems (eds Bengio S. et al.) (Curran Associates, 2018).
  24. You, J., Ying, R., Ren, X., Hamilton, W. L. & Leskovec, J. GraphRNN: generating realistic graphs with deep auto-regressive models. In International Conference on Machine Learning (2018).
  25. Dutil, F., Cohen, J. P., Weiss, M., Derevyanko, G. & Bengio, Y. Towards gene expression convolutions using gene interaction graphs. In International Conference on Machine Learning Workshop on Computational Biology (2018).
  26. Hamilton, W. L., Ying, Z. & Leskovec, J. Inductive representation learning on large graphs. In Proc. International Conference on Neural Information Processing Systems (eds Guyon, I. et al.) (Curran Associates, 2017).
  27. Fout, A., Byrd, J., Shariat, B. & Ben-Hur, A. Protein interface prediction using graph convolutional networks. In Proc. International Conference on Neural Information Processing Systems (eds Guyon, I. et al.) (Curran Associates, 2019).
  28. Ingraham, J., Garg, V. K., Barzilay, R. & Jaakkola, T. S. Generative models for graph-based protein design. In Proc. International Conference on Neural Information Processing Systems (eds Wallach, H. et al.) (Curran Associates, 2019).
  29. Jin, W., Barzilay, R. & Jaakkola, T. Hierarchical generation of molecular graphs using structural motifs. In International Conference on Machine Learning 4552–4561 (2020).
  30. Dey, T. K. & Mandal, S. Protein classification with improved topological data analysis. In 18th International Workshop on Algorithms in Bioinformatics (WABI) (eds Parida, L. & Ukkonen, E.) 6:1–6:13 (Schloss Dagstuhl–Leibniz-Zentrum fuer Informatik, 2018).
  31. Yang, C., Zhuang, P., Shi, W., Luu, A. & Li, P. Conditional structure generation through graph variational generative adversarial nets. In Proc. International Conference on Neural Information Processing Systems (eds Wallach, H. et al.) 124632 (Curran Associates, 2019).
  32. Liu, S., Grau, B., Horrocks, I. & Kostylev, E. INDIGO: GNN-based inductive knowledge graph completion using pair-wise encoding. In Proc. International Conference on Neural Information Processing Systems (eds Ranzato, S. et al.) 2034–2045 (Curran Associates, 2021).
  33. Huang, K. scGNN: scRNA-seq dropout imputation via induced hierarchical cell similarity graph. In International Conference on Machine Learning Workshop on Computational Biology (2020).
  34. Alsentzer, E., Finlayson, S. G., Li, M. M. & Zitnik, M. Subgraph neural networks. In Proc. International Conference on Neural Information Processing Systems (eds Larochelle, H. et al.) 8017–8029 (Curran Associates, 2020).
  35. Bonito, P. et al. in Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis (eds Sudre, C. H. et al.) 208–219 (Springer, 2020).
  36. Zhang, X. & Zitnik, M. GNNGuard: defending graph neural networks against adversarial attacks. In Proc. International Conference on Neural Information Processing Systems (eds Larochelle, H. et al.) 9263–9275 (Curran Associates, 2020).
  37. Xu, K. et al. Representation learning on graphs with jumping knowledge networks. In Proc. 35th International Conference on MachineLearning (PMLR, 2018).
  38. Li, M. M. & Zitnik, M. Deep contextual learners for protein networks. In International Conference on Machine Learning Workshop on Computational Biology (2021).
  39. National Research Council (US) Committee on A Framework for Developing a New Taxonomy of Disease Toward Precision Medicine: Building a Knowledge Network for Biomedical Research and a new Taxonomy of Disease (National Academies Press, 2011).

Cited by

Sorted by citations

Related

Authors
Li, M. M., Huang, K., & Zitnik, M.
Affiliations
Harvard University (US)
ORCID
Venue
Nature Biomedical Engineering
Publisher
Springer Science and Business Media LLC
Published
2022-10-31
Volume
6
Issue
12
Pages
1353-1369
ISSN
2157-846X (ISSN-L)
DOI
10.1038/s41551-022-00942-x
PMC
PMC10699434
OpenAlex
W4307717106
Semantic Scholar
167c4607722d10f350c320bc36a5612e72deba00
Cited by
294
References
218
Citations by year
2026: 66 · 2025: 85 · 2024: 93 · 2023: 42 · 2022: 8
FWCI
4.00
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Biomedical Engineering (publishedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (acceptedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University) (cc-by)
Funders
  • National Science Foundation (2030459, 2033384)
  • AstraZeneca
  • U.S. Air Force (FA8702‐15‐D‐0001)
  • Harvard Data Science Initiative, Harvard University
  • National Human Genome Research Institute (T32HG002295)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Advanced Graph Neural Networks
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
  • Physical Sciences › Computer Science › Computational Theory and Mathematics › Computational Drug Discovery Methods
Keywords
graph representation learning, network medicine, biomedical networks, graph embedding, molecular interaction networks, signaling pathways, health care system, genetic variant identification, complex traits, diagnosis and treatment, drug development
MeSH terms
Machine Learning, Delivery of Health Care, Humans

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.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

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