Teaching

BIOINF 2119 Probabilistic Methods in Artificial Intelligence (Onsite course)
Taught every spring, 2010 - 2017

This course introduces fundamental concepts and methods in artificial intelligence that are applicable to problems in biomedicine. This course is designed for students who do not necessarily have a background in computer science. The course provides the foundations in artificial intelligence methods including search (breadth-first search, depth-first search, greedy search, etc), probabilistic knowledge representation and reasoning (Bayesian networks: model, independencies, semantics, parameter estimation and inference), decision theory, and machine learning (regression, neural networks, classification trees, support vector machines, Markov models and hidden Markov models).

BIOINF 2011 Foundations of Clinical and Population Informatics (Online course)
Taught every spring, 2013 - 2014

A key goal in informatics is to represent biomedical data and knowledge in a form that can be readily used by computers, and to apply computer-based methods to biomedical data to enable computers to be useful in clinical care, in public health and in biomedical research. To address these goals, this online course introduces and provides an overview of the foundational concepts in health informatics. The course covers a variety of topics such as biomedical data and electronic health records, ontologies and standards used in biomedicine, computerized clinical decision support systems, evaluation of clinical computer systems, and computational methods that underlie clinical computer systems such as symbolic reasoning and probabilistic reasoning.

BIOINF 2011: Foundations of Clinical and Population Informatics (Onsite course)
Taught every fall, 2007 - 2011

This onsite course introduces and provides an overview of the foundational concepts of clinical and public health informatics. The course covers a variety of topics such as biomedical data and electronic health records, ontologies and standards used in biomedicine, computerized clinical decision support systems, evaluation of clinical computer systems, and computational methods that underlie clinical computer systems such as symbolic reasoning and probabilistic reasoning.

Software

The Informative Bayesian Model Selection (IBMS) is a computationally efficient method that can be applied to genome-wide data to detect both SNP-SNP interactions and interactions between two groups of SNPs (e.g., a group may consist of SNPs that map to a gene). The software for IBMS is available for download. The paper describing this method is Informative Bayesian Model Selection: A method for identifying interactions in genome-wide data and is published in Molecular BioSystems.

The backward elimination algorithm uses a kernel-based conditional dependence measure to identify the Markov blanket in a fully multivariate fashion. The software for the backward elimination algorithm is available for download at the open source code repository GitHub. The paper describing this algorithm is Markov blanket ranking using kernel-based conditional dependence measures and is published in the proceedings of NIPS 2013 Workshop on Causality: Large-scale Experiment Design and Inference of Causal Mechanisms.

The deep multiple kernel learning algorithm tunes a deep multiple kernel net by alternating optimization with the span bound. It is an attempt to extend deep learning to small sample sizes. The software for the deep multiple kernel learning algorithm is available for download at the open source code repository GitHub. The paper describing this algorithm is Deep multiple kernel learning and is published in the proceedings of the IEEE 12th International Conference on Machine Learning and Applications (ICMLA 2013).

The Modular Relief Framework (MoRF) can be used to develop novel variations of the Relief algorithm for application to genome-wide data for ranking of single nucleotide variants (SNVs). The software for MoRF is available for download at the open source code repository GitHub. The paper describing this algorithm is Application of a spatially-weighed Relief algorithm for ranking genetic predictors of disease and is published in BioData Mining.

The Model Averaged Naive Bayes (MANB) is an algorithm that predicts patient outcomes from genome-wide data by efficiently model averaging over an exponential number of naive Bayes models. The software for MANB is available for download. The paper describing this algorithm is The application of naive Bayes model averaging to predict Alzheimer's disease from genome-wide data and is published in JAMIA.

Presentations

Artificial Intelligence in Medicine MSTP Workshop, 8 March 2017. The slides are here and the citations are below: