Advanced Presentations

Statistical Interactions

BIO3 is a global leader in the epistasis area and has contributed to numerous studies and widely used methods such as Model-Based Multifactor Dimensionality Reduction (MB-MDR)1. The most recent work on this method includes its extension to the analysis of DNA-seq data (including rare variants)2, which is timely given the broad interest in high-throughput DNA sequencing. Kristel Van Steen’s lab is further improving the tool while integrating ideas from statistics and machine learning. From a practical perspective, the BIO3 lab was one of the first to publish a minimal protocol to carry out GWAIs3,4. The protocol provides guidelines and attention points for anyone involved in GWAI analysis and aims to enhance the biological relevance of GWAI findings. A more elaborate protocol, combining strengths of different methods and statistical tools, was applied to identify epistasis for Alzheimer’s Disease (AD). It led, for the first time, to replicable epistasis associated with AD using a hypothesis free screening approach5. In addition, BIO3 has overcome some of the reported obstacles related to replication and validation of epistasis findings (and sex-specificity of AD) by having adopted a sound statistical epistasis analysis protocol, supplemented with a series of experiments to show their validity and potential clinical utility. In particular, a hypothesis-free sex-stratified GWAIs for AD susceptibility was conducted, leading to the identification of a male-specific gene interaction that could be replicated in independent consortium datasets, and validated in biological experiments. This is the first contribution showing sex-specific biological epistasis in AD between genes via exhaustive genomic epistasis analysis, supported by extensive follow-up experimental work.

1. Van Lishout, F., Gadaleta, F., Moore, J.H., Wehenkel, L. & Van Steen, K. gammaMAXT: a fast multiple-testing correction algorithm. BioData Min 8, 36 (2015).
2. Fouladi, R., Bessonov, K., Van Lishout, F. & Van Steen, K. Model-Based Multifactor Dimensionality Reduction for Rare Variant Association Analysis. Hum Hered 79, 157-167 (2015).
3. Bessonov, K., Gusareva, E.S. & Van Steen, K. A cautionary note on the impact of protocol changes for genome-wide association SNP x SNP interaction studies: an example on ankylosing spondylitis. Hum Genet 134, 761-773 (2015).
4. Gusareva, E.S. & Van Steen, K. Practical aspects of genome-wide association interaction analysis. Hum Genet 133, 1343-1358 (2014).
5. Gusareva, E.S., et al. Genome-wide association interaction analysis for Alzheimer’s disease. Neurobiol Aging 35, 2436-2443 (2014).



Integrative Omics

. In 2010 think-tank activities started around a 2nd research line in parallel to the previous one: The research line on integrative analyses involves the development and implementation of integrative omics approaches for complex disease traits dissection and preventive or diagnostic medicine. BIO3’s activities have resulted in a book chapter on “Perspectives on Data Integration in Human Complex Disease”, highly appreciated conference programs for Capita Selecta in Complex Disease Analysis (CSCDA), and machine learning based (integrative) network construction methods 6-8 . Instrumental was (and still is) the COST Action BM1204 on “An integrated European platform for pancreas cancer research: from basic to clinical and public health interventions for a rare disease” (2012-2016), which is further accelerating the conception of several manuscripts related to data integration and pancreas cancer. For this Action, Kristel Van Steen has been working group leader on “Integration of omics data”.

6. Bessonov, K. & Van Steen, K. Practical aspects of gene regulatory inference via conditional inference forests from expression data. Genet Epidemiol 40, 767-778 (2016).
7. Gadaleta, F., Bessonov, K. & Van Steen, K. Integration of gene expression and methylation to unravel biological networks in glioblastoma patients. Genet Epidemiol 41, 136-144 (2017).
8. Gadaleta, F. & Van Steen, K. Discovering main genetic interactions with LABNet LAsso-based network inference. PLoS One 9, e110451 (2014).



Fine-scale Structure in Patients and Populations

In 2013, Kristel Van Steen’s earlier work on epistasis and integrative analyses led to her participation in an FP7-PEOPLE grant on “Machine Learning for Personalized Medicine (MLPM)”, as Belgian node leader and thesis supervisor for “An –omics integrated flexible framework for epistasis analysis, acknowledging interpretation capability”. One of MLPM’s work streams though was the identification of patient subtypes from heterogeneous datasets via unsupervised machine learning techniques. Finding substructure in samples (whether these refer to healthy or diseased individuals) is important to BIO3, sufficient to start a 3rd research line on the topic. Here, the main aim is to develop a multi-level data integrative tool and strategy that enables the early identification of disease-conditions related perturbations in multi-scale (-omes) measurements. Substructure in (clustering of) samples based on omics similarity should be adequately quantified and dealt with during follow-up analyses. Pilot9 and more advanced work on the theme10 have already been presented at several international conferences. This work on multi-omics profiling is expected to benefit from networking events within the COST Action CA15120 on “Open Multiscale Systems Medicine” (OpenMultiMed), for which Kristel Van Steen is a co-leader of the working group on “Multiscale Modeling”.

9. Maus, B., et al. Molecular reclassification of Crohn’s disease: a cautionary note on population stratification. PLoS One 8, e77720 (2013).
10. Chaichoompu, K., et al. IPCAPS: an R package for iterative pruning to capture population structure ( (2017).


Health Systemics: where all of the above meet …