This project aims at creating a platform to accelerate alcohol research. Machine learning algorithms are capable of analyzing high dimensional data, but lack the ability to model the more complicated features inherent in biological data. Inspired by a study of heavy drinkers that collected longitudinal neuroimaging, drinking logs, and extensive self-report and interview assessments (ABQ DrinQ), we will develop an innovative toolkit for learning and predicting alcohol use patterns. In Phase I of the project we extend support vector machines (SVMs) to handle mixed effects and thoroughly characterize our 'SVM-mix' algorithms through simulations. We then apply these methods to ABQ DrinQ to develop models of alcohol use patterns and risk factors for heavy drinking. The data, algorithms, and results are packaged into interactive notebooks for alcohol researchers to explore and compute risk in new patients. In a later Phase II, we plan to extend the platform to support additional machine learning algorithms and data features and to offer this comprehensive set of tools to early-adopters. This platform will be useful for researchers characterizing alcohol users, developing new screening tools, and developing tailored treatment programs
