Computation and Informatics in Biology and Medicine (CIBM) Seminar:
Recent advances in medicine and electronic book-keeping have greatly increased the amount of medical data available for research and clinical decision making. Electronic Health Records include information about test results, lab reports, medical images, genomics, treatments, outcomes, and family histories. Together with recent advances in data mining and machine learning, it now seems possible to realize the grand vision of predictive personalized medicine.

Statistical Relational Learning (SRL) combines the powerful formalisms of probability theory and first-order logic to handle uncertainty in large, complex problems. In this talk, I illustrate the potential of SRL to achieve an important sub-goal of predictive medicine: early detection. Specifically, I will present SRL approaches for (1) identifying young adults who are at high risk of developing Coronary Heart Disease in middle and later life, and (2) identifying the set of patients who have or will have Alzheimer's Disease by analyzing their brain MRI images. I will present a general approach for learning SRL models based on Functional-Gradient Boosting. I will adapt this algorithm for the above mentioned challenging tasks to produce state-of-the-art results in three real-world medical studies. I will outline other interesting problems in personalized medicine that we are addressing using SRL and conclude on the optimistic note that predictive personalized medicine is within reach in the near future.