Multi-view learning and deep learning for heterogeneous biological data to maintain oral health

Oral health is an integral component of overall health and well-being. The human oral cavity is the entry point for all food, the first active step of the human digestive system, and the entry point for many pathogens. Little is known about the physiological and biological processes involved in the maintenance of good oral health. Experimental data and some isolated evidences in the scientific literature suggest that biological interactions between salivary components, the oral microbiota and the host defense system actively promote the oral health status. To identify the biological interactions that underlie maintenance of oral health, one should apply a data analysis strategy that integrates multiple heterogeneous biological datasets using the best possible computational methodology. Since conventional data analysis techniques cannot handle large amount of complex biological data, one has to use appropriate ones.
Results presented in this thesis are obtained by using a set of data analysis techniques united under umbrella term "Machine learning". This thesis is aimed to study the application of single-view machine learning, multi-view machine learning and deep learning techniques to address some of the important challenges in defining oral health. The mathematical techniques covered in this thesis include biomarker selection, data clustering, multi-view data integration, and predictive modeling. Although this PhD thesis is focused on machine learning applications to oral health domain, machine learning can, in principle, help to solve problems in any biomedical domain because of the progress made in a development of multi-omics technologies and computer science.