Colloquium, Mathematical Sciences

Abstract: Cancer research increasing relies on liquid biopsies — blood samples that contain tiny fragments of tumour DNA. These measurements are low‑resolution and noisy making them a surprisingly rich source of mathematical challenges. In this talk, I will discuss how ideas from Bayesian changepoint detection, mixture models, and longitudinal inference can be used to extract meaningful signals from such imperfect data. I will talk about two methods: BayesCNA, which helps reconstructing piece-wise constant copy number profiles from extremely noisy observations; and liquidCNA, which infers the hidden underlying tumour dynamics from these profiles observed over time. Together, they can uncover the hidden evolutionary processes unfolding when a cancer gets treatment - so we can design better therapies.