Antibiotic resistance is an important public health concern that has implications in the practice of medicine worldwide. Accurately predicting resistance phenotypes from genome sequences shows great promise in promoting better use of antimicrobial agents. Machine learning is a tool of choice for this task, but the scarceness of learning examples, compared to the high dimensionality of genomic data, makes for a challenging problem.

In this talk, I will present a variety of methods that can be used to learn predictive models of antibiotic resistance from the data available in the PATRIC database. Special emphasis will be given to methods that produce interpretable models, which can be further analyzed to generate novel biological hypotheses. Notably, I will present our recent work on using set covering machines and decision trees to obtain rule-based models for this task. Moreover, I will present ongoing work on designing deep neural networks that can learn directly from whole bacterial genomes and antibiotic molecules, while highlighting the patterns (genomic and molecular) that are basis for phenotypic predictions. Finally, I will discuss limitations and future research directions.