Abstract
Human health and disease are emergent properties of a complex, nonlinear, dynamic multilevel biological system: the human body. Systems biology is a comprehensive research strategy that has the potential to understand these emergent properties holistically. It stems from advancements in medical diagnostics, “omics” data and bioinformatic computing power. It paves the way forward towards “P4 medicine” (predictive, preventive, personalised and participatory), which seeks to better intervene preventively to preserve health or therapeutically to cure diseases. In this review, we: 1) discuss the principles of systems biology; 2) elaborate on how P4 medicine has the potential to shift healthcare from reactive medicine (treatment of illness) to predict and prevent illness, in a revolution that will be personalised in nature, probabilistic in essence and participatory driven; 3) review the current state of the art of network (systems) medicine in three prevalent respiratory diseases (chronic obstructive pulmonary disease, asthma and lung cancer); and 4) outline current challenges and future goals in the field.
Abstract
Systems biology and network medicine have the potential to transform medical research and practice http://ow.ly/r3jR30hf35x
Introduction
Human health and disease are emergent properties of a complex, multilevel biological system that spans from the molecular domain to the microbiome, exposome and social levels (figure 1) [1, 2]. Ideally, therefore, if we want to intervene prophylactically to preserve health or therapeutically to cure disease, in a safe and effective way, we should understand these dynamic gene–environment interactions in greater detail. Certainly, this will not be an easy task, but the alliance of new high-throughput “omic” methodologies, novel imaging techniques and current (and future) computational power can project us forward in this endeavour and eventually facilitate the development of novel therapeutic strategies (and the repurposing of old ones) [3]. However, as wisely highlighted by one of the anonymous reviewers of this paper, to whom we are grateful: “… full understanding of complex nonlinear systems in physics and biology might not be ever possible and, fortunately, might not be even required because probabilistic decisions are (and will become) more powerful than decisions based on precise mechanistic understanding. This is a real revolution already happening in society (Google and Amazon can predict your behaviour without knowing (less understanding) you). Similarly, Artificial Intelligence (AI) will be able soon to predict the clinical course and responsiveness to intervention based on probabilities rather than on deep understanding of the system …”. We think that both concepts are actually synergistic since a more comprehensive and precise understanding of human biology (figure 1) will, no doubt, feed back to any AI platform, which will in turn provide new hypotheses to test iteratively. In any case, embracing a holistic scientific approach (as opposed to the reductionist research strategy used traditionally) for the understanding of human health and disease is a unique (and mandatory) opportunity to really move medical practice forward in the 21st century.
Multilevel layers of biological, environmental and social information ideally integrated in systems biomedicine approaches. For further explanations, see text. Reproduced and modified from [2] with permission.
In this review, we: 1) discuss the principles of systems biology, a relatively recent research strategy that leverages from omics and bioinformatics to gain a holistic understanding of complex biological systems; 2) elaborate on how this can pave the way towards the effective deployment of the so-called “P4 medicine” (predictive, preventive, personalised and participatory) [4], which can shift healthcare from treatment of illness to prediction and prevention of illness, in a revolution that will be personalised in nature, probabilistic in essence and participatory driven; 3) review the state of the art of network (systems) medicine in three prevalent respiratory diseases (chronic obstructive pulmonary disease (COPD), asthma and lung cancer); and 4) outline current challenges and future goals in the field.
Systems biology
System approaches and emergent properties
System approaches stem from the premise that separate analysis of information gathered from different elements, compartments or levels of a dynamic system (figure 1) cannot yield appropriate understanding/prediction of the global behaviour of the system (so-called emergent properties, which are implicit in nonlinear systems) nor allow to fix it if found globally away from an homeokinetic state (e.g. disease versus health), with alterations that may spread throughout various levels or compartments of the system [5]. As Macklem [5] pointed out, emergent properties arise spontaneously as self-organised order from the nonlinear interactions of the different biological components and thus the overall emergent behaviour transcends the behaviour from each part in isolation. It follows that a more holistic approach, integrating information of the interacting parts and subentities into a single mathematical representation or model, can potentially offer better clues as to the causal chain of events that leads to the apparent phenotypic manifestations and how to remedy the situation [6]. Therefore, systems biology departs from the reductionist approach followed by traditional biomedical research by integrating (rather than taking apart) different biological levels (genes, molecules, cells, organs and the environment) and mechanisms, and shares a very similar goal with integrative physiology: to better understand holistically the systemic dynamic state of individuals [7, 8]. In this context, systems biology (and systems or network medicine) is nothing more than physiology, which has always meant to be multiscale and integrative [7, 8]. The difference is that today's availability of new tools, high-throughput technologies and computing power allows, for the first time, real physiology to be performed. In essence, it is all about perspective [9]. Before “perspective” (i.e. three-dimensional) painting was “invented”, classical painting considered only two dimensions. Systems biology includes many different biological levels (dimensions) as well as the element of time dynamics. Hence, it has the potential to provide a much better definition for “the eye of the beholder” [9].
Biology as an informational science
In recent decades, faced with the biological complexity of human diseases, biomedical scientists have increasingly turned their efforts to apply high-throughput methodologies that embrace the Cartesian view that the human body is a system of formally interacting parts and that biology is an informational science. A nonexhaustive list of information sources (table 1) includes “omics” data ((epi-)genomic, transcriptomic, proteomic, metabolomic and microbiomic), single-cell analyses, phenotypic assays, extensive medical records and an endless list of environmental factors (“exposome”), such as smoking, exercise, diet and pollution, among others (figure 1). Common respiratory-specific levels of information are lung function and imaging.
Common omics data types
System representation: networks
A network (or graph) is a practical graphical representation of complex data in the context of systems approaches (figure 2), where nodes are the elements of the system under study (e.g. genes, proteins, biochemical or physiological measures, individuals or patients, among many others) and edges (or links) connect nodes that interact somehow (causality, correlation). The network(s) constructions are hypothesis driven, i.e. there is not a single, fixed, network “template”; on the contrary, they can be “custom-made”. Networks are used to make inferences regarding the emergent dynamic (spatial and temporal) behaviour of the system in response to perturbations of putative critical network elements (nodes and/or edges).
Network topology. Nodes are linked by edges. Node size represents a quantifiable node property (e.g. fold-change in two different experimental situations; this allows the inclusion of a dynamic component (i.e. time change) in the graphical representation of the network). Edge width represents the connection strength (e.g. correlation coefficient). Node colours identify different network modules. For further explanations, see text.
Diseases as network perturbations
Any disease can be viewed as a system in an abnormal state (a perturbed network) far from homeokinetic operating conditions [5], either with: 1) associated nonemergent (i.e. subclinical) alterations, or 2) observable phenotypic abnormalities (i.e. clinical symptoms) progressively departing from functional equilibrium towards partial system collapse (i.e. organ failure, etc.) or complete collapse (death). In opposition, perfect health, or wellness, can be viewed as the optimal and quantifiable state of a system in dynamic equilibrium (i.e. homeokinesis [5]).
Biological network properties
Several aspects of biomedical networks are due to their particular biological nature and must always be considered in a research setting [16]. In terms of “topology” (i.e. their spatial distribution) they are generally scale-free (as opposed to random networks). In this setting, “scale-free” means that this type of network contains many nodes with few connections and a few nodes with many links (hubs) (figure 2). This topology makes networks more robust against random perturbations [17] because of their higher modularity [18]. They are composed of loosely connected subparts (modules), which are groups of nodes highly connected internally but little to outsiders. Modules are usually coupled with specialised biological subtasks. Additionally, not all nodes are equal relative to the network structure. Central elements that are much more connected than the average are denominated “hub” nodes [19], while linkers between modules are termed “bottleneck” nodes (figure 2) [20]. Perturbations of these elements (hubs and bottlenecks) often alter the system behaviour drastically, whereas the impact of more peripheral nodes on systems behaviour (emergent properties) is often marginal. Other influential network properties with regard to the robustness of the system include “redundancy” and “degeneracy” [21]. Finally, nodes and edges may be characterised qualitatively (e.g. fold-change sign for nodes that represent gene products) or quantitatively (e.g. chemical binding constant for edges that connect drug ligands to their target molecules) (figure 2).
Medical uses
Although systems biology is best suited for experimental models of disease, it can also provide actionable and useful insights in clinical medicine [22–24]. Systems (network) medicine can lead to the identification of disease biomarkers or drug targets, both defined as key nodes whose perturbation transits the state of the biological system from health to disease or vice versa. A paradigmatic example comes from the field of cancer and the observation that the sequential use of anticancer drugs enhances cell death by rewiring apoptotic signalling networks [25].
P4 medicine
The holistic approach of systems biology discussed earlier has enabled the emergence of a new comprehensive paradigm in medicine, called P4 medicine, for predictive, preventive, personalised and participatory [4, 26–28].
From treatment to prediction and prevention
Current western medicine mostly focuses on treating diseases and symptoms when they appear. Thus, the current healthcare system organisation (and its major stakeholders, i.e. hospitals and primary care centres, pharmaceutical industry, insurance companies, policy makers, providers (e.g. physicians) and patients) is based on the provision of medication and related health products to individuals once they are sick and seeking treatment. In most countries, only a small fraction of public funds is currently devoted to prevention. Meanwhile, the burden and cost of chronic complex conditions (e.g. asthma, COPD, heart disease, stroke, cancer, type 2 diabetes, obesity and arthritis) is rising at an alarming rate [29, 30]. Hard or impossible to treat, these conditions may, however, be preventable to some (large?) extent. The hope to reach this goal (e.g. maintaining wellness) is fuelled by the growing understanding of risk factors and pathobiology of these diseases, gained (in part) as a result of the implementation of systems biomedicine approaches into research studies [31].
Personalised in nature
Since the dynamic life-long (from pre-womb to tomb [2]) interaction of genetic, environmental and social factors is what drives the physical state of individuals, and because no two individuals are biologically alike [2], the ideal preventive strategy should be tailored to each individual. In this setting, access to the individualised clouds of data, including omics data, digital medical records, and information related to the exposome, behaviour and social exposures, will be needed (and available).
Blood as a window for health assessment (liquid biopsies)
Blood stands out on the list of all the personalised data sources as it conveniently harbours dynamic critical information from many organs in the form of circulating organ-specific proteins, immune or signalling small molecules and cells (liquid biopsy). Moreover, cheap and mainstream nanosensor technology to measure these analytes longitudinally is on its way [32]. In the near future, these technologies may well serve to alert individuals in real-time to any high-risk alteration from healthy baseline measurements in order to prevent clinical complications such as organ failure, heart attacks [33], prion disease [34], liver injury [35], cancer recurrence [36], diabetes [37] or asthma attacks [38]. This scheme is expected to be especially powerful when combined with personalised genomic data, as well as other biosensors continuously tracking essential variables, such as exhaled breath [39], urine [40], imaging [41] and/or ambient pathogens or allergens [42–44].
Participatory driven
Finally, the benefits of this new P4 medicine will only be possible if patients and healthy subjects become active agents in the continuous assessment and preservation of their health. The role of health providers, both traditional (physicians, nurses, physiotherapists) and novel (genetic counsellors, behavioural coaches), will evolve to facilitate actionable information to individuals, which they can use to maintain their health [45]. Importantly, a new legal framework of rights, obligations and protections for individuals/patients and health professionals alike remains to be established and implemented. The emergence of personalised “big” data repositories raises unprecedented ethical, privacy, confidentiality, security and policy issues related to information ownership, access and management. Of note, the insurance company regulatory framework is markedly unprepared in most countries.
How to do it?
Research strategy
In principle, there are two different approaches to analyse data in this setting: “supervised” analysis based on a priori knowledge (e.g. clinical characteristics of patients) and “unsupervised” analysis (i.e. hypothesis-free). Both strategies have advantages and disadvantages, and in a sense they are complementary; their characteristics are further discussed in the Analytical complexity section.
Input data
Systems biology leverages from several omics data types. The most commonly used data types are genomics, transcriptomics, proteomics, metabolomics, epigenomics and microbiomics. Table 1 summarises their definitions, available experimental platforms, advantages/disadvantages and the bioinformatics tools needed. In each omic, data is curated, normalised and the differences between groups are usually computed using general linear models [46, 47]. We acknowledge that exposomics and imaging are missing in table 1; this is on purpose as both fields are currently developing very actively [48, 49].
Analytical complexity
Single-level analysis
A common research approach is to perform standard (supervised or unsupervised) single-level omic analysis (table 1) and then use further bioinformatics tools to facilitate the translational interpretation (table 2 and figure 3). For instance, from a list of genes/proteins of interest, in order to identify underlying biological mechanisms, functional enrichment can be performed against many databases that host annotated information on functional roles (figure 3d): Gene Ontologies of biological processes, cellular components or molecular functions [62], KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways [63], Reactome pathways [64] and gene set enrichment analysis (GSEA) [50]. Furthermore, the informational power can be augmented (“integrated”) with additional molecular interactions available from public sources with the help of dedicated software (e.g. Cytoscape plugins [65] for the STRING protein–protein association database [54]). These exponentially growing public repositories host pathway interactions (e.g. gene regulatory, metabolic or signalling) of known relationships or putative associations that come from a wide range of experiments: theoretical, animal, experimentally modified (perturbed) biological conditions, etc. Additionally, in order to obtain homogeneous groups (of variables or samples), systems medicine uses a large variety of methods for clustering data (figure 3e) [66]. The methods vary in clustering criteria, computational efficiency/speed and cluster outputs. To stratify patients, relatively simple methods such as unsupervised learning algorithms (e.g. k-means [55, 56]) are commonly applied (table 2). However, more sophisticated methods can yield hierarchically organised clusters visualised as dendrograms: bottom-up agglomerative approaches (e.g. as described by Ward [57]) are preferred when clusters are expected to contain few observations, while top-down divisive approaches (e.g. DIANA (divisive analysis clustering) [58]) are more suited for estimating large numbers of clusters [67].
Widely used tools to facilitate biomedical interpretation from omics analysis
Summary of bioinformatic methodologies currently available. a) Correlation (e.g. Pearson/Spearman) network constructed from omics data. b) Weighted gene coexpression networks analysis (WGCNA) methodology. c) Bayesian networks approach. d) Gene set enrichment with gene set enrichment analysis (GSEA). e) k-means clustering. For further explanations, see text.
Finally, networks can be used to infer the structure of the data (table 2). Co-expression, co-occurrence or similarity networks can be built either across all study samples or separately for each medical condition that is to be compared (figure 3a). Many methods have been devised and are continuously under active development for building these networks. The simplest approach is to compute all gene–pair correlations with conventional coexpression measures (Pearson/Spearman/Kendall, mutual information [60]) (figure 3a), but more complex statistical procedures exist that cater for the specific features of biological systems. An extensively used procedure is termed weighted gene coexpression networks analysis (WGCNA) (figure 3b) [59, 68], which has the added value of clustering genes into nonoverlapping modules and correlating them with any imputed (usually clinical) characteristic. WGCNA can be complemented with functional enrichment analysis.
Multilevel analysis: the true revolution
Although some studies have and will continue to work successfully on a single omic level, recent decades have seen an ever-increasing body of work where several distinct omics datasets, including also other biological or clinical levels, are analysed conjointly using multiscale integrative methods such as SNF (similarity network fusion) [69]. This combination of levels has the potential to provide researchers with simultaneous information from several compartments of the biological system of interest, thus facilitating the modelling of the dynamic nonlinear relationships that characterise emergent properties (phenotypes) and complex diseases. Accordingly, this strategy would be able to provide more power to identify groups of patients affected with the same pathobiological mechanism or more power to probabilistically model (without understanding) the health versus disease states. The main multiscale analytical tools described to date are summarised in table 3. The “supervised” methods can be grouped mostly into either network-based, machine learning or multistep approaches [86], while the “unsupervised” can be further classified as based primarily on networks, Bayesian approaches or matrix factorisation (table 3).
Analytical strategies for integrative multiomics analysis
Current applications of systems approaches in respiratory medicine
The pathogenesis of most common respiratory diseases is complex and largely undefined from a precise pathobiological point of view. Chronic respiratory conditions, such as asthma or COPD, are still diagnosed (and treated) based on respiratory symptoms and traditional lung function measures, but they are highly heterogeneous and often overlap. In fact, they are the end result of complex genetic and environmental interplays that are yet to be explicitly modelled. This poorly defined characterisation of the basic disease mechanisms results in nonspecific, mostly symptom-driven treatment options, or lack thereof, that may eventually be able to slow the progression of these diseases in fortunate, responsive patients.
Systems biology and network medicine approaches are being put forth in an effort to palliate this painful lack of knowledge and understanding by tackling two fundamental and interrelated matters: 1) as in other biomedical fields such as cancer, a novel classification (i.e. “taxonomy”) of chronic airway diseases is needed, based not on clinical presentation (i.e. “phenotypes”) but instead either on the underlying biological mechanisms (i.e. “endotypes”) when characterised or resulting directly from data-driven probabilistic clustering of patients data; and 2) a more precise patient stratification that can be transferred to distinct and personalised preventive or therapeutic prognosis as well as improved prognosis (i.e. P4 medicine) is also needed, as recently highlighted in a review focused on biological therapies for airway diseases [87].
COPD
COPD is a heterogeneous disease with pulmonary and extrapulmonary manifestations [88], and variable response to pharmacological treatment [89], suggesting that the condition affects several distinct biological pathways. To characterise this heterogeneity at the molecular level, several studies have already used a number of different systems approaches. 1) WGCNA and GSEA showed that a molecular signature composed of gene modules related to B-cell activity, NK-cell activity or viral infection cellular markers might be detectable in peripheral blood months following COPD exacerbations [90]. 2) Xue et al. [91] used other network-centric procedures to reveal an unexpected loss of inflammatory signature in COPD patients, as well as an activation-independent core signature for human and murine macrophages. 3) Glass et al. [92] used the network inference analysis PANDA (Passing Attributes between Networks for Data Assimilation) [93], designed for improved integration of individual with public datasets, and discovered network rewiring of lymphocyte activation signalling circuits in a known gene variant implicated in COPD by genome-wide association studies. 4) Faner et al. [94] unravelled differences in the molecular pathogenesis of emphysema and bronchiolitis by performing correlation network analysis of lung transcriptomics on COPD patients. They found that B-cell-related genes were significantly enriched in emphysema (compared with COPD patients without emphysema), paving the way for differential therapeutic research on inflammatory pathways of the adaptive immune response. 5) Two COPD studies demonstrated the utility of unsupervised k-means clustering by identifying robust cluster associations with clinical characteristics and known COPD genetic variants [95, 96]. 6) Very recently, Ross et al. [97] introduced a new Bayesian method for COPD subtyping. They applied it to the COPDGene cohort and identified nine different patient subgroups with distinct disease progression trajectories. Of note, Ross et al. [97] prove that their sophisticated model has a better predictive capacity than multivariate ordinary least squares regression analysis.
Asthma
Several international consortia have already applied systems biology and network medicine approaches to asthma research. 1) ADEPT (Airways Disease Endotyping for Personalised Therapeutics) and U-BIOPRED (Unbiased Biomarkers for the Prediction of Respiratory Disease Outcome Consortium) have, for instance, applied a clustering algorithm to two independent asthma cohorts on a small set of easily measurable clinical variables and successfully defined four longitudinally stable clusters of patients with distinct clinical and biomarker profiles (from blood, sputum and airway data) [98]. 2) Kuo et al. [99] recently reported three novel molecular phenotypes of asthma in the U-BIOPRED cohort by analysing sputum cell transcriptomics in asthmatic and nonasthmatic subjects. They applied hierarchical clustering of differentially expressed genes as well as gene set variation analysis, gene–protein coexpression and pathway enrichment analysis. 3) Sharma et al. [100] used network-based tools to analyse the predictive value of the asthma interactome, and characterised high-impact pathways central to the disease heterogeneity and drug response. 4) Qiu et al. [101] used PANDA on participants of the Childhood Asthma Management Program cohort to assess the differential connectivity between the gene regulatory network of good responders to inhaled corticosteroids versus that of poor responders. The method allowed them to integrate their dataset with public data interactions of genes, transcription factors and proteins, and eventually implicate several network hubs and transcription factors (as well as regulatory rewiring) in the heterogeneity of drug treatment effects. Specifically, the differential network topology of good responders versus that of poor responders revealed enriched corticosteroid-induced pro-apoptosis pathways in the former and anti-apoptosis pathways in the latter, as well as key regulatory transcription factors (hubs) that drove differential downstream gene expression in the two groups.
Lung cancer
Lung cancer is the leading cause of cancer death in the world. Lung cancer is highly heterogeneous genetically because of a high mutation rate, as well as extremely complex since it comprises a disparate subset of diseases with distinct and possibly overlapping pathobiologies that share a common phenotypic manifestation. Smoking is a core shared risk factor for COPD and lung cancer; up to 65–70% of lung cancer patients suffer both lung cancer and COPD [102, 103]. So far, no single satisfactory circulating (i.e. liquid biopsy) tumour marker has been properly validated, but recently a panel of six tumour markers showed a very high specificity and sensitivity in patients referred to a tertiary hospital because of the clinical suspicion of lung cancer [104, 105]. Given that inherited genetic variants play a significant role in lung cancer development [106], but contribute little to risk estimates of classical predictive statistical models [107–109], it is hoped that systems biology approaches will allow the comparison multilevel high-throughput omics data between tumour and normal tissue, and facilitate the identification of early diagnostic lung cancer biomarkers. WGCNA has already been used successfully in lung cancer research. 1) Tang et al. [110] related the gene expression profile of lung squamous cell carcinoma with five differentially expressed long noncoding RNAs that could help in prognosis evaluation. Their gene signature was statistically associated with overall survival in important clinical subsets (stage I, epidermal growth factor (EGFR) wild-type and EGFR mutant). 2) Tian et al. [111] analysed coexpression networks and protein–protein interactions of data available in public repositories (The Cancer Genome Atlas, KEGG and Gene Ontology).
What's next? Future challenges
For the successful development and implementation of systems biology and network medicine approaches in respiratory medicine, several challenges need to be faced and eventually solved.
Technical challenges
In any clinical study, only a fraction of the biological variability is captured (and therefore analysed) due to technical limitations (and cost) of the experimental tools available. The development of new experimental tools (e.g. high-throughput next-generation sequencing, mass spectrometry-based flow cytometry or real-time molecular imaging) will generate new information but, at the same time, massive amounts of (big) data that will have to be adequately handled, analysed and interpreted [112–114]. In this context, Riekeberg and Powers [115] recently reviewed the methodological advancements and successful applications of metabolomics, one the newest omic fields.
However, research would benefit not only from measuring “more” relevant variables, but also from estimating with better precision those variables already determined in the context of a more complete definition of reference and pathological ranges (that vary in time, across individuals and biological codeterminants) [116]. Of the variability supposedly present in experimental data, these currently unaccounted factors and batch effects should not be underrated since they can partly explain the general difficulty to replicate scientific findings in the biomedical field, of which respiratory biomedicine is not exempt.
Computational challenges
Computational methodologies and programming analytical tools are being constantly refined as they translate advancements from complementary areas such as AI and information science. However, challenges and difficulties remain. For instance, in differential expression (omics) analysis, one of the main difficulties is to accurately separate the biological signal from technical noise. As risk factors of complex diseases are polygenic (individualised genes have little independent effects), batch effects and technical heterogeneity are difficult to separate from protein, gene and epigenetic perturbations causing the disease. Computational analysis partly palliates this difficulty by integrating enormous amounts of data into models or network representations. However, there is not a single bioinformatic approach for the task and each has its own best-use cases, advantages and drawbacks. Regrettably, this lack of standard biostatistical and algorithmic procedures may create a lot of uncertainty as to the validity of the results and usually cannot fully guarantee reproducibility [117]. It is thus extremely important that bioinformatic researchers assess the sensitivity and optimal network thresholds of their implementations. Thus, replication and experimental validation of results must remain a research priority.
Drug discovery challenges
Network (poly) pharmacology differs from conventional drug discovery strategies by providing a powerful rationale for target identification that is based on the analysis of disease-specific biological networks. This novel paradigm consists of administering simultaneously multiple small molecules to target several biochemical network nodes in an attempt to “re-engineer” the network into its normal and healthy dynamic structure [118]. It has the potential to overcome the two major obstacles hindering the field: efficacy and toxicity. This new approach still requires considerable methodological developments. Proof-of-concept was obtained by combining methods as diverse as network analysis, text mining, molecular docking data and the STRING database [54] to integrate data from network pharmacology and metabolomics [119].
Healthcare system, educational and economic challenges
The milestones required for systems medicine to become a reality go far beyond mere scientific and technological progress. The structure of the healthcare system needs to substantially adapt to operate with multidisciplinary teams [113] of traditional (physicians, epidemiologists, computational biologists, IT specialists, statisticians) and new roles (genetic counsellor [45], behavioural coach, specialised educators), which cannot function without specific omics data storage facilities, diagnosis centres, standard analytical pipelines and managerial frameworks. Furthermore, systems biology and P4 medicine require specific education (via higher education degrees, complementary formations for hospital personnel or genomics training programmes [120]) since there is a growing gap between the amount of data generated by basic scientists and the clinical expertise available to analyse, interpret and translate it into clinical practice. Finally, because of the high cost of overcoming these challenges, there is a significant risk that individuals in developing countries will not benefit from this health revolution [121], unless guided by the appropriate political efforts from the international community and an informed public opinion [122–124]. Even within the richest nations, equitable access to P4 medicine is not guaranteed, and low-income individuals may not be able to afford the unsubsidised cost of omics data integration, diagnostic and clinical care [125, 126].
Conclusions
Human health and disease are emergent properties of a complex, multilevel and dynamic system: the human body. Systems biology and network medicine are comprehensive research strategies that have the potential to understand the emergent properties of the system holistically. By doing so, they are paving the way for a radical shift in medical practice that is evolving from a reactive proposition to a predictive, preventive, personalised and participatory (P4) approach. Respiratory medicine is in fact already contributing significantly to this change by leading the field of data-driven management [98, 99] as well as by applying multilevel network analysis to a variety of clinical conditions [127].
Acknowledgements
The authors thank the two anonymous reviewers of our manuscript for their very helpful and constructive criticisms and suggestions.
Footnotes
Number 4 in the Series “Personalised medicine in respiratory diseases” Edited by Renaud Louis and Nicolas Roche
Previous articles in this series: No. 1: Chung KF. Personalised medicine in asthma: time for action. Eur Respir Rev 2017; 26: 170064. No. 2: Bonsignore MR, Suarez Giron MC, Marrone O, et al. Personalised medicine in sleep respiratory disorders: focus on obstructive sleep apnoea diagnosis and treatment. Eur Respir Rev 2017; 26: 170069. No. 3: Mascaux C, Tomasini P, Greillier L, et al. Personalised medicine for nonsmall cell lung cancer. Eur Respir Rev 2017; 26: 170066.
Support statement: This work was supported, in part, by Instituto de Salud Carlos III (PI12/01117, PI15/00799, CP16/00039), Recercaixa-2012 (AA084096), SEPAR (PI065/2013, PI192/2012, PI68/2015), AGAUR FI-DGR 2016 PhD grant and AstraZeneca foundation COPD Young Researcher Award. CERCA Programme/Generalitat de Catalunya. This work was developed at the building Centre de Recerca Biomèdica Cellex, Barcelona, Spain. Funding information for this article has been deposited with the Crossref Funder Registry.
Conflict of interest: None declared.
Provenance: Commissioned article, peer reviewed.
- Received September 28, 2017.
- Accepted November 30, 2017.
- Copyright ©ERS 2018.
ERR articles are open access and distributed under the terms of the Creative Commons Attribution Non-Commercial Licence 4.0.
References
