Machine Learning and Gut Microbiome for Breast Cancer Screening

The gut microbiome can provide key information regarding host health and disruptions to the microbiome have been associated with various diseases including breast cancer. A critical factor in BC survival is early detection, when spread is limited. Consequently, analyzing the gut to screen for BC provides an opportunity to introduce a non-invasive, low cost yet reliable screening tool. In this study, we identify key microbial biomarkers from fecal samples and leverage supervised learning techniques to develop a model for BC detection. We conduct experiments across nine algorithms and four feature selection and engineering techniques on 3 splits of data: Invasive-Control, Benign-Control, and Invasive-Benign. Analyzing the results, we highlight the most informative biomarkers for each category. Our best performing models achieve Area under Curve (AUC) score of 0.829, 0.943 and 0.920 for Invasive-Control, Benign-Control and Invasive-Benign respectively.