Digital Twins for Friedreich’s Ataxia: A Sequence-to-Sequence Model of Disease Progression

Abstract

Friedreich’s ataxia is a debilitating genetic disease typically diagnosed in childhood. It is characterized by loss of coordination, mobility, and independence, but like many other rare diseases, there is currently no cure. This thesis focuses on the potential for digital twin technologies to transform healthcare and presents a transformer-based approach to the synthetic control problem to investigate two key applications: (1) control simulations for in silico clinical trials and (2) synthetic interventions to support clinical decision making. While synthetic control methods traditionally use time-agnostic weights to capture similarities across the unit space, we propose a novel approach that is flexible across space and time. We find that this approach significantly outperforms state of the art synthetic control methods on both synthetic data and a Friedreich’s ataxia dataset. Furthermore, we demonstrate how our model can be directly applied to major prediction problems in medicine and create the first Friedreich’s ataxia disease progression model of its kind.