PreMedIT

The goal of precision medicine is to tailor treatment according to patients’ characteristics, by reliably figuring out which patient will benefit from a given treatment and when to initiate such a treatment. For this, one must identify markers predicting dynamically the response of a given patient to a treatment, which is a delicate task as such markers can be of diverse natures, in large numbers, and have complex interactions. Recent advances in machine learning are opening promising prospects, to determine which treatments are best suited to a patient. However, the data encountered in medicine pose crucial challenges: the datasets can be small due to the inclusion criteria, they can be sparse, and they may contain numerous categorical variables. We conjecture that integrating information about the geometric and topological structure of the data into the machine learning pipelines will help address these challenges.