Poster sessions

Name
Zhenwei Shi

Research group/department
MAASTRO Clinic

Abstract
Purpose: Radiation-induced lung disease (RILD), such as dyspnea, is a risk for patients receiving high-dose thoracic irradiation. This study is a TRIPOD Type 4 validation of previously-published lung toxicity models via secondary analysis of esophageal cancer SCOPE1 trial data. We quantify the predictive performance of these two models for predicting dyspnea 6 months after high-dose chemo-radiotherapy for primary esophageal cancer.

Material and methods: We tested the performance of the earlier models using baseline, treatment and follow-up data on 258 esophageal cancer patients in the UK enrolled into the SCOPE1 multi-centre trial. The adverse event of interest was dyspnea ≥ Grade 2 within 6 months of the end of radiotherapy. External validation was performed using an automated, decentralized approach, without exchange of individual patient data.

Results: Out of 258 patients with esophageal cancer in SCOPE1 trial data, 38 patients developed radiation-induced dyspnea (≥ Grade 2) within 6 months of the end of radiotherapy. The discrimination performance of the models in esophageal cancer patients treated with high-dose external beam radiotherapy was moderate, AUC of 0.68 (95% CI 0.55 – 0.76) and 0.70 (95% CI 0.58 -0.77), respectively. The curves and AUCs derived by distributed learning were identical to the results from validation on a local host.

Conclusion: We have external validated previously published dyspnea models using an esophageal cancer dataset. Prediction performance was not statistically different from previous training and validation sets. The distributed learning approach gave the same answer as local validation, but is feasible without accessing a validation site’s individual patients-level data.

Name
Olin Janssen

Research group/department
Alzheimer Center Limburg, department of Cognitive Neuropsychology and Clinical Neuroscience

Abstract
Objective: Alzheimer’s disease (AD) is the most common cause of dementia without a cure so far. To better understand the clinical and pathophysiological mechanisms of AD and to identify which treatments would be best suited for different patient groups, a disease model is needed that encompasses all the available evidence. The aim of the European IMI ROADMAP project is to model the progression of AD across the full disease spectrum and to lay the foundation for a European-wide Real-World Evidence (RWE) platform on AD.

Methods: RWE and Randomised Controlled Trial (RCT) data sources relevant to AD will be identified, extracted, harmonised, integrated and analysed. Examples of data sources are cohort studies, national registries, RCT placebo data, health care registries, electronic medical records and data from general practitioners. Key outcome measures across stakeholder groups will be identified, and guidelines for combining different RWE data sources in AD will be developed.

Innovation: Patients, caregivers, pharmaceutical companies, academia, regulatory authorities, health technology assessment bodies and reimbursement agencies are all involved in ROADMAP and collaborate in creating an integrated RWE data environment. By looking at health data in the widest sense through combining population and dementia-related data across Europe, healthcare challenges in AD can be addressed.

Anticipated results: Modelling the progression of AD by combining different data sources will improve understanding of the impact of the disease on for example quality of life, treatment costs and health resource utilization. Ultimately, the model allows for prediction of real-life disease progression and improved clinical decision making.   
 

Name
Ilona Liesenborghs

Research group/department
Maastricht Centre of Systems Biology (MaCSBio)

Abstract
Introduction: Glaucoma is a potentially blinding eye disease. Trabecular meshwork (TM) tissue plays an important role, however, the pathogenic mechanism is not clear yet. We performed in silico pathway analysis based on gene expression datasets of healthy human TM cells. Identification of the active pathways in this tissue contributes to a better insight in the molecular processes and may provide candidate target genes for new treatment options.

Methods: Gene Expression Omnibus and ArrayExpress were searched for microarray datasets. 16 potential datasets were identified. Quality control and pre-processing were performed with ArrayAnalysis.org and divergent samples were removed from the datasets. Thereafter, the datasets were integrated with each other. Pathway analysis and visualization was performed with PathVisio. Ubiquitous pathways were identified and excluded based on data from Wikipathways TissueAnalyzer. The top 25% genes with the highest expression value were used as cut-off for pathway analysis. The top 20 Z-score ranked pathways were selected as these are likely the most relevant TM active pathways.

Results: During the analysis, some problems appeared: incomplete information about the datasets and donors, non-uniform data, unclear which processes already had been performed. For now, we completed the workflow for 11 datasets. Several of the identified pathways have been associated with processes in the TM, for example the TGF-β signaling pathway (WP366) and the miRNA targets in extracellular matrix and membrane receptor (WP2911).

Conclusions: Combining data, although challenging, can provide a more complete and better insight into the gene expression profiles and molecular processes of healthy tissues and diseases.

Name
Leonard Wee

Research group/department
Department of Radiation Oncology (MAASTRO), GROW – School for Oncology and Development Biology

Abstract
Main objective: To develop a healthcare “big data” prediction model for absolute risk of death (by any cause) following treatment for esophageal cancer. The architecture semantically links data extracted from patients’ medical images and clinical records at two independent treatment clinics (Maastricht, The Netherlands and Cardiff, Wales). Our objective is to train and validate a machine learning model via distributed learning across these clinics.

Significance: Data fragmentation, privacy regulations and language barriers are the main challenges to address on the field of medical imaging. Transformation of healthcare “big-data” following FAIR principles is a crucial step to render the data accessible to machine learning and artificial intelligence, in combination with machine assistance for relevant treatment outcome predictions.

Methodology: Image-based biomarkers (radiomics) were automatically extracted from medical images of 85 patients, taken prior to treatment. Clinical and radiomics features were parsed into RDF using the Radiation Oncology and Radiomics ontologies. A privacy-preserving distributed learning approach was used for independent validation of the prognostic model for risk of death.
 
Innovation: A novel quantitative approach was used to automatically extract structured data from medical images instead of human-operator qualitative/descriptive approaches. Furthermore, statistical learning and model validation is feasible on distributed, semantically-annotated FAIR resources without the need to exchange data or reveal individual patient identities.

Results/future benefits: The validation of the results is in progress, aiming to evaluate the radiomics features that are significantly associated with the overall survival, establishing a distributed learning workflow of radiomics and clinical features used in daily clinical practice.

Name
Elisa Cirillo

Research group/department
BiGCaT- Bioinformatics department

Abstract
Problem: Single Nucleotide Polymorphisms (SNPs) from the Genome Wide Association Studies (GWAS) are challenging to interpret in the disease context. In particular, the non-coding SNPs are not systematically addressed in follow up analysis.

Importance: The non-coding SNPs are the preponderant output of GWAS, and they contribute to the majority of the heritability of a complex disease. Their single effect is small and mainly regulatory, but all together are predicted to explain the regulation mechanisms of a disease.

Method: A previous Body Mass Index (BMI) GWAS study was analyzed. Different levels of complexity such as: SNPs, genes and pathways data were combined towards the understanding of the obese condition. The function of non-coding SNPs was elucidated by using the epigenetic data and the expression quantitative traits loci (eQTLs) from adipose tissue, skeletal muscle, liver and pancreas. The first data type allows to define if a variant is located in an active chromatin region, and the second shows if the variant can influence gene expression in specific tissue.
Innovation: The data integration was performed with network analysis, and for the best of our knowledge this is the first study that present a combination of GWAS, epigenetic, eQTLs and pathways data all in one network.

Results: An in silico overview of the epigenetic activity of BMI non-coding variants and their influence on the gene expression, is presented in four tissue specific networks. Relationships within SNPs, genes and pathways are also represented, enabling the effect interpretation at the process level, of variants not defined before. The networks will be publicly available for further exploration and interpretation. 

Name
Matthijs Cluitmans

Research group/department
CARIM School for Cardiovascular Diseases, cardiology department

Abstract
Background: Abnormalities in the electrical propagation of the heart beat are an accepted substrate for life-threatening cardiac arrhythmias. However, subtle-but-potentially-life-threatening abnormalities might not be detectable from the clinical electrocardiogram. Noninvasive imaging of electrical activation and recovery directly at the heart surface in combination with computational modelling may provide patient-specific insights. 

Objective: To apply noninvasive electrocardiographic imaging (ECGI) in patients to detect abnormal gradients of electrical recovery directly at the heart surface, and to investigate the use of a computational model of electrophysiology (EP) to obtain more insight in the patient-specific interactions that then may lead to arrhythmias.

Methods & Results: In a 47-year old female with unexpected out-of-hospital cardiac arrest, ECGI was applied to investigate the electrical substrate and abnormally triggered beats. ECGI exposed abnormal gradients of electrical recovery near the origin of the abnormally triggered beats, suggesting an interaction between these triggers and (clinically concealed) substrate.
A computational model of cardiac EP was developed and validated with invasive experimental data. The computer model was then used to mimic the recovery gradients and trigger origins found in the patient, to study wave front propagation under different conditions. This revealed that only the combined presence of 1) recovery gradients, and 2) an early (short-coupled) triggered beat could result in life-threatening arrhythmias.

Conclusion: Integration of advanced imaging and simulations of cardiac electrophysiology helps to infer patient-specific arrhythmia mechanisms. In patients with latent substrate, this approach could potentially help risk-stratify patients for sudden cardiac arrest, who might benefit from protective therapy.

Name
Alberto Traverso

Research group/department
Maastro Clinic

Abstract
Main objectives: A) To build an infrastructure to transform patients’ available information following FAIR (Findable Accessible Interoperable Reusable) principles. B) To integrate the infrastructure with a workflow to compute clinically relevant prediction models by using mentioned data. C) To produce the output of previous workflow as semantic-labelled data to allow reproducibility and experiments’ benchmarking.

So far, most of clinical prediction models are based on clinical trial data. Despite the high quality of these data (well structured, absence of missing values), they represent only a small portion of ‘big’ clinical patients data. Therefore, models based on clinical trials data could suffer from concerns about generalizability. To avoid these issues, models should be ‘trained’ on ‘big’ clinical data within a multi-center approach, verifying their reproducibility and generalizability.

Ontologies and Semantic Web were used to transform ‘big’ clinical data into FAIR data. Transformed data was then used to construct multivariable models using a distributed learning and multi-center approach. An innovative aspect is that the output of the models were then exported as ontology-labelled data. In this way, different methodologies (e.g. prediction algorithms) / results can easily be compared and experiments become reproducible. In addition, a standardization effort (Image Biomarker Standardization Initiative) for computing quantitative image features has started and has led to the definition of consensual methods in the community.

We developed three dedicated ontologies for clinical data. We created an infrastructure for distributed learning. Future benefits of FAIR quantitative imaging include the validation of prediction models on a large scale.

Name
Friederike Ehrhart

Research group/department
Department of Bioinformatics

Abstract
The current environment of rare disease data resembles a landscape with silos.  Apart from the data being scattered in different silos some of these silos also serve different purposes, including: 1) patient data registries 2) genetic data repositories which is increasing since  whole exome sequencing is used as clinical standard for diagnosis of rare disorders 3) genotype-phenotype databases and 4) databases which store general information about genes, proteins, metabolites and their interactions.

The current problems are that these data silos are interesting for researchers and clinicians on different levels but should be interoperable for optimal use while guaranteeing the patients anonymity.

Solutions for these problems are currently being developed. There are numerous patient registries and at least 12 genotype-phenotype databases with different degrees of FAIRness. Nomenclature of genetic data according to guidelines from the Human Genome Variation Society (HGVS) is by now commonly accepted although some databases deal with different identifiers e.g. reference SNP. Another emerging approach is the use of nanopublications describe genetic variation data. This RDF based approach has the advantage that it is possible to include not only the variation but also the provenance and information about the reliability and, due to the  flexibility of RDF graphs, allows to link more information e.g. a phenotype description. An ELIXIR organised Bring Your Own Data (BYOD) event held at Maastricht University started to link genetic variation data with phenotype description and resulted in an Elixir funded implementation study to shed some more light on the problems and validate possible solutions. 
 

Name
Samar Tareen

Research group/department
Maastricht Centre for Systems Biology (MaCSBio)

Abstract
Obesity is identified as a major risk factor for multiple chronic diseases and diet induced weight loss is the preferred method to counter the effects of obesity. Adipose tissue is highly affected in weight loss, but little is known about the core underlying molecular mechanisms. We used time series gene expression data of subcutaneous adipose tissue from a weight loss study to decipher the said mechanisms by studying the biological processes being affected by the gene expression patterns.

The dataset consists of two separate diets for weight loss: a low calorie diet at 1250 kcal/day and a very low calorie diet at 500 kcal/day. The constructed pipeline starts with the normalisation of the expression data followed by filtering for background noise. The filtered dataset was then used to find differentially expressed genes, which were then used to generate correlation networks, one for each diet, based on the correlation of the expression patterns of the genes over the time points within the respective diets. The two correlation networks were then overlapped to find the conserved correlated expression patterns between the two diets. Topological clustering was performed on the overlap network to find clusters of closely correlating gene expression patterns. Gene ontology enrichment of the overlap network as a whole and of the clusters individually yields an additional layer of process-level detail of the subcutaneous adipose tissue, as well as new areas to explore for pathway- and process-level crosstalk regulating the adipose tissue response to diet induced weight loss.

Name
Nicole Vogelzangs

Research group/department
Epidemiology

Abstract
The obesity epidemic is a growing public health threat, but obesity-related cardiometabolic risk is very heterogeneous. Recent evidence indicates that insulin resistance (IR) in obesity may develop independently in different organs, representing different etiologies towards cardiometabolic diseases. The aim of this study is to investigate whether distinct metabolic profiles are associated with IR in muscle or liver.

This study includes 659 overweight/obese (BMI≥27) non-diabetic participants (18-65 years; 63% women) of the European DiOGenes project. Muscle insulin sensitivity index (MISI) and hepatic insulin resistance index (HIRI), were derived from a 5-point oral glucose tolerance test. 18 plasma metabolites were identified and quantified by NMR. Results were validated in a sample of 349 non-diabetic subjects (BMI≥27; 40-65 years; 54% women) from the Maastricht Study.

Results suggest that early stages of muscle and liver IR are characterized by clearly distinct metabolic profiles. Muscle IR is not associated with profound plasma metabolite alterations. In contrast, hepatic IR associates with increased levels of branched-chain and other amino acids (higher valine, leucine, isoleucine, alanine, tyrosine, proline; in women: oxo-isovaleric acid, hydroxyisobutyrate) and a metabolic profile suggestive of increased carbohydrate catabolism and gluconeogenesis (higher glucose, lactate; in women: lower glycine), and decreased hepatic fat oxidation (lower levels of ketone bodies: acetoacetate, 3-OH-butyrate; in women: higher triglycerides). More pronounced associations in women suggests hepatic insulin resistance might have a larger impact in women than in men. Results may guide the development of better tailored sex- and tissue-specific IR interventions in the prevention of cardiometabolic disease.

Name
Kirill Tumanov

Research group/department
Department of Data Science and Knowledge Engineering

Abstract
Not all people are enjoying the ability to interact with the outer world in a seamless way – some have deficiencies which leave them only few ways of expressing themselves (e.g. locked-in patients). In this research we would like to allow humans to interact with their environment through a brain-robot interface. Using this interface, humans should be able not only to express themselves, but also to achieve practical benefits in their daily routines, which might at present be carried out by the caregivers. Specifically, we design and build a non-invasive, affordable, and easy to use interface to control a mobile physical robot through mental activations using functional Near-InfraRed Spectroscopy (fNIRS).

The robot platform used currently allows its user to navigate to a certain spot of a known environment (e.g. person’s house) and perform simple actions (e.g. patrol entrances, engage in conversations through telepresence). In the future we plan to enrich the repertoire of the available robot tasks by employing different types of robots (e.g. robot manipulators, drones), so that the target audience does not feel themselves limited in the ways they interact with the world.

The novelty of the research comes from methodological side – we design the methods of brain signal processing and classification, as well as look into the robotic aspects to develop the reliable and responsive brain-robot interface. In the course of the research we perform experimental trials with healthy volunteers to validate our methods and demonstrate prospects of the technology.

Name
Alexander Koch

Research group/department
Department of Pathology

Abstract
For decades, scientists have been searching for cancer biomarkers to improve the diagnosis and treatment of cancer. This resource-intensive endeavor has resulted in thousands of biomarker publications, but translation of these markers into clinical practice hardly takes place. Estimations by Poste (2011) and Kern (2012) put the number of published markers that are used in the clinic below one percent. This failure to translate biomedical findings into clinical applications has been termed the translational research valley of death (Butler, 2008).

If we are to cross this biomedical valley of death for DNA methylation markers, we have to increase the scientific quality and reproducibility of biomarker research, reduce the number of markers lost in translation, and accelerate the development of clinically useful markers. This will require a coordinated effort from all stakeholders: scientists, funding organizations, scientific journals, private partners, and patients.

We have devised a strategy to tackle these issues and improve the reliability, efficiency and translation of cancer DNA methylation markers, while at the same time promoting data sharing. At the heart of this strategy is the construction of a database of all published markers. This unique database would provide researchers with a valuable resource where they can find and evaluate existing markers or, in collaboration with private partners, have a marker experimentally validated. Using the collected data, we also plan to develop a reporting standard for DNA methylation markers. The ultimate goal of our efforts is to offer cancer patients more and better biomarkers.

Name
Shauna O'Donovan

Research group/department
Maastricht Centre for Systems Biology (MaCSBio)

Abstract
Disturbances in adipose tissue fatty acid dynamics due to insulin resistance have been reported as a major link between obesity and type 2 diabetes mellitus or cardiovascular disease. Currently several mathematical models exist which aim to describe postprandial fatty acid dynamics across the adipose tissue. However, due to limitations in available data it has been difficult to determine exact parameter values and to test the assumptions upon which these models have been constructed. Using adipose tissue flux data coupled with palmitate tracer data we aim to validate and modify, where necessary, the various model terms in order to fully capture the dynamics of the system.

As part of the YoYo Study sixteen individuals (BMI: 28-35kg/m2) underwent a high fat mixed-meal challenge test containing a stable isotope palmitate tracer at baseline and following a five week period of weight stabilisation after weight loss. Samples were collected at -30,0,60,120,180,240, and 300 mins from an arterialised hand vein and a vein draining the abdominal subcutaneous adipose tissue. The YoYo Study A-V data was used to compare and validate model terms and assumptions from three existing models. Terms describing LPL lipolysis, LPL fractional spill-over and glycerol handling the existing were modified and the glucose and NEFA uptake models extended in order to more accurately capture the dynamics of the system. Comparison of parameter sets obtained from fitting the resulting computational model to baseline and following weight stabilisation data indicate a reduction in the delay of insulin stimulation of several reactions following weight loss.
 

Name
Michelle Moerel

Research group/department
Maastricht Centre for Systems Biology (MaCSBio)

Abstract
Objective: To explore sound processing throughout the cortical depth of the human auditory cortex.

Significance: Results will serve to next explore changes in sound processing with attention and learning.

Methods: We acquired high-resolution MRI data while volunteers listened to natural sounds. Cortical responses were analyzed with two computational models that represented different hypotheses on sound processing inside and outside the primary auditory cortex (PAC). The simple tonotopy model described sound processing by the cortical frequency preference, while the joint modulation model hypothesized frequency-specific tuning to spectrotemporal modulations. Topographic maps were made by color coding voxels according to the trained joint modulation model.

Innovation: The layers of the neocortex each have a unique anatomical connectivity and functional role. Their exploration in the human brain, however, has been severely restricted by the limited spatial resolution of non-invasive measurement techniques. Here we use ultra-high field fMRI at 7 Tesla to make the investigation of this small spatial scale feasible.

Results: We observed that while deep and middle PAC layers were well represented by a simple frequency model, neuronal populations in superficial PAC displayed an increase in processing complexity better represented by the joint modulation model. This increased processing complexity was present throughout cortical depths in the non-PAC. These results suggest that a relevant transformation in sound processing takes place between the thalamo-recipient middle PAC layers and superficial PAC. This transformation may be a first computational step towards sound abstraction and perception, serving to form an increasingly more complex representation of the physical input.
 

Name
Ana Casas

Research group/department
Department of Pharmacology & Personalised Medicine

Abstract
Based on non-hypothesis-driven approaches genetic evidence suggests that diseases are interrelated differently to our current organ-based ontology. In fact, common effector mechanisms, when affected or triggered seem to produce pathophenotypes in diverse organs or co-morbidities. This will lead eventually to a revised disease nomenclature and opens up entirely new approaches for diagnosis and treatment.

In this context, we noted that a common cardiovascular target, the cGMP-forming Fe(II) haem protein, soluble guanylate cyclase (sGC), appears to be situated in a common mechanism network that is prominently relevant to stroke. Ischemic stroke is the second leading cause of death worldwide and the leading cause of disability. Despite this high medical need only a single drug is available but due to its side-effects 85% of all patients are excluded from treatment. Upon middle cerebral artery occlusion sGC protein and nitric oxide-stimulated activity in the ischemic hemisphere were dramatically down-regulated leading to a high proportion of oxidized and/or haem-free apo-sGC activity. Pharmacological targeting of apo-sGC in vitro under oxygen and glucose deprivation conveyed strong neuroprotection via ERK/CREB signalling. In vivo, post-stroke apo-sGC activation by two distinct members of this compound class augmented cerebral blood-flow whilst leaving systemic blood pressure unaffected, reduced infarct size and increased survival. Different apo-sGC activators are in advanced stages of clinical development for different cardiovascular indications. Systems biology and network medicine and our preliminary target validation suggest that they should be urgently tested for repurposing as first-in-class, mechanism-based neuroprotective drugs in stroke.

Name
Martin Wetzels

Research group/department
Department of Marketing and SCM

Abstract
Based on non-hypothesis-driven approaches genetic evidence suggests that diseases are interrelated differently to our current organ-based ontology. In fact, common effector mechanisms, when affected or triggered seem to produce pathophenotypes in diverse organs or co-morbidities. This will lead eventually to a revised disease nomenclature and opens up entirely new approaches for diagnosis and treatment.

In this context, we noted that a common cardiovascular target, the cGMP-forming Fe(II) haem protein, soluble guanylate cyclase (sGC), appears to be situated in a common mechanism network that is prominently relevant to stroke. Ischemic stroke is the second leading cause of death worldwide and the leading cause of disability. Despite this high medical need only a single drug is available but due to its side-effects 85% of all patients are excluded from treatment. Upon middle cerebral artery occlusion sGC protein and nitric oxide-stimulated activity in the ischemic hemisphere were dramatically down-regulated leading to a high proportion of oxidized and/or haem-free apo-sGC activity. Pharmacological targeting of apo-sGC in vitro under oxygen and glucose deprivation conveyed strong neuroprotection via ERK/CREB signalling. In vivo, post-stroke apo-sGC activation by two distinct members of this compound class augmented cerebral blood-flow whilst leaving systemic blood pressure unaffected, reduced infarct size and increased survival. Different apo-sGC activators are in advanced stages of clinical development for different cardiovascular indications. Systems biology and network medicine and our preliminary target validation suggest that they should be urgently tested for repurposing as first-in-class, mechanism-based neuroprotective drugs in stroke.

Name
Nalan Basturk

Research group/department
Quantitative Economics

Abstract
The main objective of AQuARIUM is to foster and engage Maastricht University (UM) researchers from different disciplines, working on concrete societally relevant projects, with the aim of making significant advances in research. AQuARIUM is a project-based learning program to develop collaboration and deeper understanding of data science amongst UM researchers.

We will tackle two problems regarding multidisciplinary data science research. First, different areas use similar data science methods, but the applications, subsequent problems and solutions differ across fields. Second, different areas develop different methods but the applicability of these methods in different contexts requires a deep understanding of the methods and supporting theories. This problem will be addressed in problem-based ideation sessions, working on selected projects, with the goal of bridging and disseminating knowledge across fields.

For 2018, the first edition of AQuARIUM takes place at the School of Business and Economics in collaboration with the D3M research theme, and will be centered around the following projects: Food policy design: balancing commercial interests and public health; Optimizing engagement: an online gaming case study; Data-driven HRM: maximizing the effectiveness of PhD student allocation; Data-driven policy evaluation: capturing human behaviour and expectations through text analytics.

The project-based ideation sessions are aimed at developing innovative research on societal problems. The basis is an in-depth presentation of quantitative methods from mathematics, statistics and machine-learning followed by project-specific collaboration. The ideation sessions have as end of the year goal the creation of top quality research papers and creation of project proposals for international funding. In the future, we aim to extend AQuARIUM projects to UM level.

Name
Chang Sun

Research group/department
Institute of Data Science at Maastricht University

Abstract
Health data is vertically partitioned across many different sources including personal data vaults, hospitals, insurers, municipalities, schools and others. The privacy and sensitivity of health data makes it more difficult to share and combine those data.

In this project, we will handle the challenges from scientific, technical, and social-legal-ethical issues to build a scalable technical and governance framework which can combine access-restricted data from multiple entities in a privacy-preserving manner. The area of application is to examine socio- economic factors that contribute to diabetes.

Challenges

• Scientific: How to combine and learn from access-restricted FAIR health and socioeconomic data across entities in a privacy-preserving manner?
• Technical: How to analyze Big Data in health from multiple resources without sharing and observing the original data?
• Legal-ethical: How to define and underpin the responsible use of Big Data in health?