Machine learning-assisted precision medicine for smoking cessation

Machine learning-assisted precision medicine for smoking cessation
Center for Human Genomics and Precision Medicine Seed Grant
John Curtin (PI), Tim Baker (Co-I), Megan Piper (Co-I), Jerry Zhu (Co-I), James Li (Co-I)
Direct Costs: $50,000
Status: Funded. 09/01/2020 – 08/31/2021

Precision medicine offers an approach whereby treatments are selected to optimize their therapeutic benefit for individual patients based on the patient’s unique characteristics (e.g., traits, history, environmental factors, genetics).

Although precision medicine approaches for guiding treatment selection have been pursued for decades, progress has been hindered by 1) the use of traditional analytic techniques that cannot account for the complex (high-dimensional) relationships between individual differences and treatment success necessary for accurate prospective prediction; and 2) the failure to include valuable genetic features alongside and in interaction with psychological, demographic, and environmental features. Contemporary machine learning approaches are well-suited to accommodate high-dimensional arrays of features, to explore genetic and non-genetic features simultaneously, and to extract reliable prediction signals that generalize robustly to new samples of patients.

The proposed project seeks to apply machine learning approaches and incorporate human genomic features into the precision medicine paradigm for cigarette smokers to guide treatment selection among several first-line smoking cessation medications.

Preliminary data suggest that there is clinical utility in this approach, such that using model-based treatment selection yields clinically meaningful increases in predicted treatment success compared to treatment success when treatments are randomly assigned. This project will build on preliminary work by increasing the sample size, considering additional candidate statistical algorithms, and incorporating polygenic scores to develop a precision medicine model and demonstrate its ability to guide treatment selection for individuals quitting smoking.