Work package 8
Combining exposome, multiomic, and imaging data using artificial intelligence to identify phenotypes of groups at-risk for developing lung diseases
Multiomics analyses as applied in WP6 on longitudinally collected serum, urine, breath and fecal samples, as well as exposome data (WP5), imaging data (WP7) will be combined and analyzed in this work package.
It has become clear that the complexity and heterogeneity of lung diseases cannot be encompassed by single biomarkers and that the exposome is an important factor in the development of lung diseases. Furthermore, multiomics integration has already proven to allow for a drastic improve of unsupervised molecular group classification of COPD20. Therefore, a combination of the data collected in the other work packages will allow us to maximize the information we can retrieve from this data. This ‘big data’ approach introduces the need for sophisticated bio-informatics and bio-statistical tools. In this work package we aim to improve the classification of persons at-risk to develop lung diseases.
Conventional statistical methods will be used for correlations of single layer -omics biomarkers between the subjects that have early signs of disease versus disease-free subjects despite a similar high-risk exposome at start. Artificial Intelligence (AI) incorporating approaches, including deep learning techniques such as artificial (convolutional) neural networks and decision tree algorithms will be used for supervised analyses. These will be set up to separate the group of subjects that developed early sings of disease and the group of subjects that did not, despite a similar exposure at start. The goal is to detect patterns in the combined datasets that are able to discriminate between those groups. Within the subjects that developed (first signs of) disease, unsupervised techniques will provide further insight in the existence of different clusters, reflecting different phenotypes, differences in the exposome and (longitudinal) differences in progression of disease.
The deliverable of this work package is the identification of phenotypes of groups that are at-risk to develop lung diseases. Information about these phenotypes will subsequently be implemented in unravelling the mechanism underlying lung injury and –repair (WP10) and to optimize prevention strategies (WP9, WP11).
Reference
20. Li C-X et al. Integration of multi-omics datasets enables molecular classification of COPD. Eur Respir J. 2018;51:1701930.