Machine Learning Methodologies for Uncovering the Mechanisms of Complex Diseases

Identifying the mechanisms associated with complex human diseases is one of the main challenges of human genetics and computational medicine. With every technical advance such as the ability to sequence whole genomes and transcriptomes faster and cheaper, there is the promise of finally discovering the root causes of complex genetic diseases and leading the way for efficient personalized medicine. That promise have not been fulfilled yet despite the technological advances and large datasets gathered. All the variations discovered to be statistically associated with a disease can only account for a small proportion of the genetic component of the disease. Given the complexity of the problem, we need to move beyond the state-of-the-art statistical tools and use advanced machine learning techniques that can better fit the complexity of the problem. In this thesis, we propose machine-learning methods that are designed specifically for the biological problem in order to increase the power of identifying mechanisms of complex diseases. We present approaches that can integrate multiple types of data when each dataset contains only a part of the disease mechanism and we incorporate as much biological knowledge and annotations as possible within our framework. We tackle the problem by simultaneously considering all potential sources of aberrations that can lead to a disease and show that we can find more complex mechanisms explaining the disease in a larger proportion of the patients. Our approaches also provide guidelines to investigate the likely root causes for individual patients opening the way to personalized treatment.Ph.D