Integrative omics to characterize and classify pulmonary vascular disease

Accumulating evidence supports the utility of omics as a window into disease pathobiology or as an adjunct to aid precision clinical phenotyping¹. Recent advances in high-throughput biotechnologies and analytical methodologies has enabled widespread utilization of genomics, transcriptomics, proteomics, and metabolomics (collectively referred to as ‘omics’) for deep phenotyping of patients who are at-risk for or with established pulmonary vascular disease (Fig. 1). Omics analyses are typically performed using blood samples, but the testing platforms may also be used to assay a variety of biospecimens, including disease tissue, urine, saliva among others. Results from omics studies collectively have the capacity to provide an all-encompassing view of the molecules important for cellular or tissue structure and function.

Figure 1. Omics and precision phenotyping.

Omics platforms that analyze genomics, the transcriptome, proteomics, and metabolomics provide information on DNA, RNA, proteins, and metabolites, respectively at the cellular, tissue, or patient level. When combined with clinical data, omics data allows for precision phenotyping and clustering of patient populations. Even with clustering of like phenotypes, individuals have unique omics signatures and some heterogeneity remains within clusters. E1, exon1; I, intron; EHR, electronic health record; Rx, prescription.

The pulmonary vascular disease cohort is particularly amenable to using omics to interrogate disease pathobiology as pulmonary vascular and right ventricular tissue is not readily available. As symptomatic patients tend to present late in their clinical course, peripheral (or central) circulating omics signatures function as a liquid biopsy of the pulmonary vasculature or right ventricle, and, thereby, can provide a snapshot of a patient’s disease status. Furthermore, results from omics studies, when paired with clinical and outcome data, can provide diagnostic as well as prognostic information and identify disease biomarkers.

To date, the majority of defining omics studies in pulmonary vascular disease have been performed in patients with pulmonary arterial hypertension (PAH) and studied a single type of omics. More recently, integrating data across omics platforms has emerged as the next step to inform (endo)phenotyping in pulmonary vascular and other diseases. This, in turn, has created big datasets and a shift to favor analytical methodologies that provide optimal and actionable data outputs. Herein, we discuss omics studies in the context of pulmonary vascular disease and unbiased analytical approaches to illustrate the use of integrated omics datasets to enhance endophenotyping and clinical phenotyping in patients with pulmonary vascular diseases.

Genomics and pulmonary vascular disease

There was early recognition that PAH was associated occasionally with a strong family history of both PAH and sudden death in young family members, suggesting unrecognized PAH². Concerted international efforts led to the discovery that loss-of-function mutations in the bone morphogenetic protein receptor type 2 (BMPR2) gene, a member of the transforming growth factor-β (TGFβ) superfamily, underlie approximately 80% of heritable PAH^3–5. Interestingly, an important feature of this familial form of PAH is incomplete penetrance, such that only ~20% of persons with a BMPR2 mutation ultimately develop the PAH phenotype. Subsequently, it was also recognized that BMPR2 mutations are present in about 10–15% of patients with idiopathic PAH, perhaps representing de novo mutations in the gene or the first known affected person within a predisposed family⁶. BMPR2 mutations, however, did not account for all heritable PAH and other early studies identified mutations in the TGFβ signaling pathway such as activin receptor-like kinase (ALK1)^7,8 endoglin (ENG)⁹, and SMAD8¹⁰ related to PAH.

Advances in next generation sequencing have facilitated new discoveries of genetic causes of PAH in more diverse pathways adding to the concept that there is molecular heterogeneity in the disease. Whole exome and whole genome sequencing have identified mutations in SMAD1, SMAD4, SMAD9, and caveolin 1 (CAV1)¹¹ as well as several other novel pathways including the potassium channel Two Pore Domain Channel Subfamily K Member 3 (KCNK3)¹², growth differentiation factor (GDF2)¹³, T-box transcription factor 4 (TBX4)¹⁴, ATPase polyamine transporter (ATP13A3)¹³, and aquaporin-1 (AQP1)¹³. Eukaryotic translation initiation factor-2-α kinase 4 (EIF2AK4) has also recently been associated with pulmonary veno-occlusive disease and pulmonary capillary hemangiomatosis^15,16 (Fig. 2). In general, these are rare variants that are inherited in an autosomal recessive fashion. Recent efforts have focused on identification of common genetic variants that confer increased risk for PAH and/or adverse clinical outcomes. These studies identified a locus overlapping the HLA-DPB1 gene¹⁷ and polymorphisms in SOX17¹⁷ and Cerebellin (CBLN2)¹⁸ as highly likely candidates.

Figure 2. Genetic causes of PH.

Advances in high throughput sequencing have allowed for new genetic causes of PH to be identified. Beyond bone morphogenetic protein receptor 2 (BMPR2) and related pathway members (endoglin and ACVRL1) and the ligand BMP9, these include mutations in SMAD9, and membrane-related proteins caveolin 1 (CAV1), the potassium channel Two Pore Domain Channel Subfamily K Member 3 (KCNK3), aquaporin-1 (AQP1), and the sulfonylurea receptor 1 protein of the ATP-sensitive potassium channel (ABCC8). Other variants have been identified in the transcription factors T-box transcription factor 4 (TBX4) and SRY-box transcription factor 17 (SOX17) as well as the endosomal ATPase polyamine transporter (ATP13A3). Eukaryotic translation initiation factor-2-α kinase 4 (EIF2AK4) has also recently been associated with pulmonary veno-occlusive disease and pulmonary capillary hemangiomatosis.

From Southgate L, Machado RD, Graf S, Morrell NW. Molecular genetic framework underlying pulmonary arterial hypertension. Nat Rev Cardiol. 2020;17(2):85–95; with permission.

Despite our understanding of the genetic causes of heritable PAH, in idiopathic PAH, as well as other forms of PH, the majority of cases are without a clear genetic cause, although these areas continue to be actively investigated. In addition, while the most progress has been made on identifying mutations and rare variants associated with PAH, there is a growing literature on the potential role of epigenetic modifications, such as DNA methylation¹⁹ and chromatin ultrastructural changes^20,21 as mediators of pulmonary vascular disease.

Transcriptomics to understand Pulmonary Vascular Disease

With the recognition that gene mutations or modifications may explain only a minority of PH cases, studies have been extended to examine RNA expression patterns and microRNAs in pulmonary vascular disease. A potential challenge to RNA expression studies is, however, the changing nature of RNA patterns. RNA expression, both in blood and tissue, is necessarily dynamic and acutely responsive to exposures and environmental changes. Therefore, findings from analyses of RNA are likely to reflect both the state of pulmonary vascular disease etiology and recent internal and external environmental exposures.

Early studies found clear differences in RNA expression patterns between the lungs of PAH patients compared to controls^22,32. Similar findings were observed when peripheral blood mononuclear cells were analyzed. The advantage of studying these cells is that they are relatively easy to obtain, offer the convenience of repeated sampling, and potentially reflect the underlying vascular pathology ²³. RNA sequencing has also proven useful to identify novel mechanisms through which endothelial BMPR2 suppression may affect pulmonary vascular disease: when BMPR2 was decreased, there were differences in transcripts related to DNA damage and repair mechanisms, nitric oxide signaling, and growth factor signaling, all of which have been implicated in PAH disease biology ²⁴. Comparison of lung tissue transcripts from 58 patients with idiopathic PAH and 25 control subjects revealed differential expression of 1,140 transcripts. Using gene set enrichment analysis, it was discovered that estrogen receptor 1, interleukin-10 receptor A, tumor necrosis factor-α, and colony-stimulating factor 3 were upstream regulators of the transcripts. Analysis of biological pathways revealed that 16 were significantly “rewired” in PAH as compared to controls, suggesting that in pulmonary vascular disease, there were newly formed or restructured gene interdependencies, which has implications for identifying druggable targets in the disease. Among the interesting findings from these analyses was the identification of negative regulation of WNT-related processes as a potential mediator of sex-based differences in PAH²⁵. More recently, RNA sequencing of whole blood samples obtained from patients with idiopathic, heritable, or drug-induced PAH revealed a panel of 25 transcripts that could distinguish PAH patients from control subjects and was associated with disease severity as well as survival. Further analysis revealed an association between lower SMAD5 levels in whole blood and susceptibility to PAH²⁶. This advantage of the aforementioned study is the use of whole blood, indicating that transcriptomics doesn’t require hard to obtain biospecimens, e.g. lung or heart tissue, to be informative. Other studies of RNAs have also focused on noncoding RNAs, including microRNAs and long non-coding RNAs (reviewed in ²⁷), which have been shown to modulate pulmonary vascular disease.

Novel integration of other omics platforms with transcriptomics, such as chromatin and interaction profiling, has also been utilized to reveal how epigenetic factors affect the pulmonary artery endothelium from patients with PAH. These analyses found that in PAH, there was extensive remodeling of acetylation at H3K27ac, which marks an increase in the activation of transcription, without any evidence of differences in promoters or gene expression. Cell-type specific gene regulatory network analysis identified transcription factors that were active in PAH compared to controls and regulated 1,880 genes in the network. This gene set was enriched for pathways related to endothelial-to-mesenchymal transition and response to growth factors ²⁸. The value of this integrated omics approach is the additional information revealed by combining data identified a novel endothelial phenotype transition in pulmonary vascular disease, which may have been overlooked had a single omics platform been studied.

Proteomic profiling of pulmonary vascular disease

Proteomics provides a comprehensive assessment of protein expression, abundance, and post-translational modifications with functional implications. In this manner, proteomics serves as an endophenotypic intermediate between genomics and metabolomics. Over the past 15 years, proteomic studies in pulmonary vascular disease have been performed using varied biospecimens, different disease cohorts and comparator groups, as well as technologies, including mass spectrometry, aptamer-based assay, or antibody-based assays. This has resulted in the identification of many disease-related biomarkers with credible biological relevance, although this heterogeneity has made it difficult to compare between smaller studies and results from large-scale clinical studies are limited. Nonetheless, proteomics has identified new disease pathways and putative biomarkers.

Results from proteomic studies have implicated inflammation and activation of immune signaling in pulmonary hypertension (PH). An early proteomic analysis that utilized plasma from patients with idiopathic PAH found a 4-fold increase in complement 4a des Arg, a component of complement, in PAH patients compared to healthy controls ²⁹. The biologic plausibility of complement activation as a biomarker in PH was confirmed recently in pre-clinical models of PH and the pulmonary vasculature of patients with PAH as well as in patients with congenital heart disease-related PH ^30,31. Another study utilized unsupervised machine learning to define 4 distinct clusters of immune phenotypes from directed proteomic analysis of chemokines, cytokines, and pro-inflammatory mediators. These clusters did not segregate by underlying disease etiology, were prognostic, and were associated with five-year survival rates ³².

Subsequent proteomic analyses performed using PAH lung tissue explants revealed differences in 25 proteins, including chloride intracellular channel 4, receptor for advanced glycation end products, and periostin. These proteins are related to cell proliferation, migration, and metabolism, all of which are cellular processes that have been linked to PAH³³. Similar studies performed using blood outgrowth endothelial cells from patients with hereditary PAH identified differences in levels of 22 proteins, including translationally controlled tumor protein (TCTP), which is related to cell proliferation and apoptosis-resistance in cancer³⁴. Global proteomics and phosphoproteomics of pulmonary artery endothelial cells isolated from patients with PAH or healthy controls found differential expression of proteins involved in metabolic pathways, nitric oxide generation, and oxidant stress, suggesting mitochondrial dysfunction³⁵. Taken together, these studies share common themes that relate pulmonary vascular disease to inflammation and immune-mediated responses, proliferation and apoptosis, and dysregulation of metabolism.

As PH has been associated with exercise intolerance, a proteomic study of skeletal muscle tissue was done to investigate potential etiologies of this phenomenon. This analyses revealed that key proteins related to mitochondrial function and metabolism were downregulated in the skeletal muscle. This finding supported many of the observed abnormalities in the skeletal muscle tissue, including abnormal mitochondrial morphology, decreased oxidative phosphorylation and increased expression of glycolytic proteins³⁶.

Results from proteomic studies have also identified panels of proteins that function as prognostic biomarkers. In one multicenter study, 20 proteins were identified in the plasma proteome of patients with idiopathic or heritable PAH that differentiated survivors from nonsurvivors. A panel of 9 proteins, including proteins related to myocardial stress, vascular remodeling, metabolism, inflammation and immunity, thrombosis, and iron regulation, all mediators of PH that were confirmed in preclinical or clinical studies, was found to be prognostic. Interestingly, this panel was predictive independent of plasma NT-proBNP levels and was shown to improve the prognostic significance of the REVEAL score³⁷.

While the majority of proteomic studies have been performed using biospecimens from patients with PAH, there is emerging evidence from other patient populations at risk for, or with established pulmonary vascular disease. In patients with heart failure with preserved ejection fraction (HFpEF), relevant clinical, laboratory, and echocardiographic data were utilized to identify six distinct clinical phenogroups that differed in all-cause mortality and heart failure hospitalizations at follow-up. Analysis of plasma proteomics found that 15 proteins that included inflammatory and cardiovascular proteins differed between the phenogroups³⁸. In a larger study that included 877 individuals in the discovery cohort, proteomics identified 38 proteins that were associated with incident heart failure and replicated in a validation cohort. These proteins collectively identified 4 biologically relevant pathways associated with progression to heart failure: inflammation and apoptosis, extracellular matrix remodeling, blood pressure regulation, and metabolism³⁹. Although these studies were not limited to include only patients with left heart failure and pulmonary vascular disease, they identify many of the same pathways implicated in PAH, suggesting commonalities between the underlying diseases and progression to pulmonary vascular disease.

Proteomic studies in CTEPH are also limited. To gain insight into pulmonary vascular remodeling in CTEPH, proteomic analysis of endarterectomized samples was compared with a control sample that was a composite of cultured human pulmonary artery endothelial, smooth muscle, and fibroblast cells. The analysis revealed 679 endarterectomy-specific proteins that were related to several key biological processes, including complement and coagulation cascade pathways and extracellular matrix receptor interactions, among others. Although interesting, the control sample was comprised of cultured cells and didn’t include platelets, which make the significance of the findings unclear⁴⁰.

Metabolomics as a snapshot of pulmonary vascular disease and right ventricular dysfunction

Perturbations in the metabolome has been a central theme in studies of the etiology of PH^41,42. Modern capacity to study metabolomics more broadly using mass spectrometry, high-performance liquid-phase chromatography, or nuclear magnetic resonance spectroscopy has accelerated recognition of the number of metabolic pathways that are altered in PH. Similar to RNA transcripts, metabolomic profiles are affected by exposures, hormones, fasting, exercise, and circadian rhythms such that the metabolome has the propensity to change rapidly ^43,44.

Studies of endothelial cell metabolomics have identified disruption of several pathways, including an increase in glycolysis, a reduction in glucose oxidation, and a decrease in fatty acid oxidation^42,45,46. These findings were confirmed in PH patients that underwent a hyperglycemic clamp. This revealed an impaired insulin response to hyperglycemia in the PH patients compared to controls with evidence of increased skeletal muscle insulin sensitivity. Metabolomics profiling found elevated ketones and lipid oxidation in PH with fatty acids, acylcarnitines, insulin sensitivity, and ketones correlating with disease severity. Thus, in PH, control of glucose is limited and lipid and ketone metabolism favored ⁴⁷. These metabolomic insights are beginning to translate to trials of new therapies for PAH including dichloroacetate⁴⁸, metformin (NCT0361745), and ranolizine^49,50.

There is also emerging research on metabolomics in other patient populations with or susceptible to pulmonary vascular disease. In heart failure patients, metabolomics profiling has both diagnostic and prognostic value. Early studies demonstrated that patients with heart failure could be discriminated from healthy controls based on levels of 4 metabolites: histidine, spermidine, phenylalanine, and phosphatidylcholine C34:4⁵¹. These metabolites, however, were not replicated in another study that identified different metabolites that discriminated heart failure from healthy controls⁵². Other metabolomic studies identified a panel of metabolites, which included increased levels of hydroxyleucine/hydroxyisoleucine and decreased levels of dihydroxydocosatrenoic acid and hydroxyisoleucine as predictors of incident heart failure⁵³. Differences in the study design and the populations studied likely contributed to the heterogeneity in the metabolite profiles.

Perhaps one of the most distinct pulmonary vascular disease populations, and, therefore, one that is well suited for metabolomics studies is chronic thromboembolic pulmonary hypertension (CTEPH). Recently, plasma metabolomic profiles comparing CTEPH, idiopathic PAH and healthy controls found that CTEPH patients have evidence of aberrant lipid metabolism with increased lipolysis and fatty acid oxidation⁵⁴. In the future, we will likely see metabolomic profiling applied to broader populations of patients with pulmonary vascular disease where the use of transpulmonary and transcardiac gradients of metabolomics to understand metabolism across these organ systems will have unique value⁵⁵.

Cardiomyocytes are highly metabolically active and in the normal state prefer fatty acid oxidation to glucose oxidation. In both rodent models and human disease, the failing right ventricle has reduced fatty acid oxidation and enhanced glucose metabolism^56–60. While readily demonstrated by [¹⁸F] FDG PET studies⁴², these changes have been demonstrated ex vivo in the RV using metabolomic studies as well^56,57,61. There have also been studies correlating RV function with less traditionally recognized cardiomyocyte metabolic pathways. For instance, using global metabolomic profiles of pulmonary vascular and RV function at rest and with exercise, indoleamine 2,3-dioxygenase (IDO)-dependent tryptophan metabolites have been shown to correlate well with baseline RV function⁶². Presently, it is unknown how, if at all, to modulate metabolism in the RV to improve outcomes in pulmonary vascular disease.

Perhaps most relevant to clinical care are studies of metabolomic data from plasma to improve clinical predictors of outcomes ⁶³. For instance, metabolomic profiling identified 20 circulating metabolites that differentiated PAH patients from both healthy and disease controls as well as 36 metabolites that were confirmed as independent prognostic markers, including transfer RNA-specific modified nucleosides, TCA cycle intermediates and fatty acid acylcarnitines. Importantly, several of the metabolites tracked with disease severity over time such that correction of metabolite levels correlated with improved clinical outcomes⁶³. These data show the potential role of metabolomics in clinical care and assessing risk as well as response to therapy. Like studies of the plasma transcriptome, the plasma metabolome likely changes in the short- and long-term, which makes it more suited as an indicator of response to therapy than, for instance, DNA variants that are unchanged over time. While not ready for clinical use presently, we anticipate that future clinical trials will likely incorporate studies of metabolomics to define treatment responses, particularly in the context of metabolic interventions.

Multi-omics phenotyping, big data, and novel methodologies for integrating data

As multi-omics phenotyping gains traction, consideration of how to approach and analyze these big datasets is warranted. As many prevalent diseases are multifactorial and represent a composite of endophenotypes, it is more likely that results from a multi-omics approach will have broadly applicable prognostic or therapeutic implications in pulmonary vascular disease and PH. Consideration of using a multi-omics study design is especially important when lifestyle or environmental exposures may modify the phenotype as occurs in in pulmonary vascular disease. For this reason, multi-omics profiling will be pursued to understand all forms of pulmonary vascular disease, as outlined in the omics plan for the NHLBI Pulmonary Vascular Disease Phenomics Study (PVDOMICS) study⁶⁴. Other challenges associated with multi-omics phenotyping include defining a relevant sample size, heterogeneity of the disease and control populations studied, significant heterogeneity in the methodologies used to measure and analyze omics that may limit cross-study applicability; and, the necessity for a validation cohort or biological confirmation in preclinical models⁶⁵.

Analytical methods for big datasets generated by multi-omics first require data integration at a high level. There are two approaches that are often undertaken to analyze the combined outputs from genomics, transcriptomics, proteomic and metabolomic datasets. The first method is post-analysis data integration. Here, each of the omics datasets is analyzed individually to generate results and integration of the varied datasets is performed after the initial analysis. This methodology allows for selection of data from each of the platforms that is deemed relevant to pulmonary vascular disease, which is then integrated or networked together for further analysis. In contrast, the second method utilizes computational tools to integrate all the omics data into a single big dataset and, once integrated, analysis performed on the entire dataset. This approach has the added advantage of reducing bias⁶⁶.

Analyzing omics big datasets has necessitated the introduction of newer methodologies, such as machine learning and systems or network analysis. Machine learning is a form of artificial intelligence and can be supervised or unsupervised. When machine learning is supervised, the computer algorithm is provided with a training dataset that includes the omics data and how it is linked to an outcome of interest, such as clinical worsening in pulmonary vascular disease patients or mortality. This allows for the development of a rule that links omics with outcomes and can be used to predict outcomes when a new omics dataset is introduced. In unsupervised machine learning, none of this information is provided forcing the algorithm to test all possibilities that link the input (omics data) to the output (clinical outcomes)⁶⁷. This methodology allows for prediction, prognostication, and clustering to reveal new pulmonary vascular disease phenotypes.

Network analysis is a second newer analytical method that allows for unbiased discovery of previously unknown relationships between genes, proteins, and metabolites in an integrated omics dataset. Network analysis maps the components of the integrated omics studies based on known interactions. This creates the network structure where each component is referred to as a node, connections between nodes are referred to as edges, and highly connected nodes are referred to as hubs. In networks, interactions between nodes are unlikely to occur by chance, suggesting that functional relationships exist among nodes ^68,69. Those genes, proteins, or metabolites that are essential tend to encode hubs that reside at the network’s center while disease-related genes, proteins, or metabolites frequently reside at the network’s periphery ^68–70. The network architecture may also explain or predict a phenotype. For instance, perturbation of a single node is likely to have many repercussions throughout a network and affect other connected nodes ^68,71. Groups of nodes in the same neighborhood within the network tend to assemble into modules that are identified using unbiased clustering algorithms ^68,70. These modules may contain a group of nodes that share a similar function, or contain disease-related genes. Nodes can also overlap with other modules and reveal novel relationships between disease- and non-disease factors that would be important to understand, particularly in the context of druggable targets and potential side effects ^68,69. In some cases, where the “directionality” of the interactions between nodes is known, flux analyses can be performed to identify key regulatory points⁷². In pulmonary vascular disease, network analysis has facilitated our understanding about microRNA regulation of hypoxia, inflammation, and TGFβ signaling pathways⁷³ as well as identify relevant nodes in key pathways that can are drug targets or may benefit from repurposing of existing drugs ^65,67.

Conclusion

It is now recognized that deep phenotyping of patients with pulmonary vascular disease in the precision medicine era requires the inclusion of omics studies. Omics testing can be performed using blood sampled from various compartments, relevant disease tissue if available, or other biospecimens. Omics has the potential to provide a comprehensive assessment of the genes, transcripts, proteins, and metabolites that govern the structure, function, and metabolism of pulmonary vascular cells at timepoints ranging from prior to disease inception to advanced states of disease. Omics have been utilized to identify putative biomarkers and have clinical relevance for diagnostic and prognostic capabilities in PH. More recently, it was realized that it is feasible to integrate multi-omics data across data types and platforms to provide even more granular information about the pulmonary vascular disease phenotype and more robust results. When coupled with clinical data, integrated multi-omics data is the future of precision phenotyping in pulmonary vascular disease.

Figure 3. Approach to integrated data analysis.

Multi-omics testing generates big datasets that that can be analyzed separately or integrated to create a novel mega dataset. There are two approaches that have been employed to integrate multi-omics data. A) Initially, results from the omics platforms are analyzed separately. Once the data is available, relevant data from each of the omics platforms can be selected and combined into a new larger dataset for new analyses. In this example, data from each of the omics platforms identifies significant molecules (BMPR2, SMAD5, TCTP and Taurine). These and other data are then selected and integrated into the new dataset. Analysis may then reveal a new biomarker that was not previously identified in any of the individual omics studies or can be used to generate a network that identifies factors that group together and form a module (red nodes). B) A second method involves integration of the multi-omics data prior to any analysis. Once combined, the mega dataset is analyzed for discovery of new relevant disease biomarkers as well as diagnostic or prognostic factors.

Synopsis:

Advances in high throughput biotechnologies have facilitated omics profiling, a key component of precision phenotyping, in patients with pulmonary vascular disease. Omics provides comprehensive information pertaining to genes, transcripts, proteins, and metabolites. The resulting omics big datasets may be integrated for more robust results and are amenable to analysis using machine learning or newer analytical methodologies, such as network analysis. Results from fully integrated multi-omics datasets combined with clinical data are poised to provide novel insight into pulmonary vascular disease as well as diagnose the presence of disease and prognosticate outcomes.

Key points:

1. In patients with pulmonary vascular disease, results from omics studies can be used to understand disease pathobiology, diagnose disease in at risk patients, select medications and therapeutics, or prognosticate clinical outcomes.

2. The majority of omics studies performed to date in the pulmonary vascular disease field have focused on a single omics platform (i.e., genomics, transcriptomics, proteomics, metabolomics) and does not integrate different types of data.

3. Newer methodologies to analyze integrated multi-omics big datasets, such as network analysis, have the potential to identify previously unrecognized disease relationships, identify new drug targets, or find targets that can be treated by repurposing approved drugs.