Transfer Learning from Well-Curated to Less-Resourced Populations with HIV
Proceedings of the 5th Machine Learning for Healthcare Conference, PMLR 126:589-609, 2020.
In Europe and North America, more homogeneous virus types and the relatively high availability of sequencing technologies have helped transform HIV from a life-threatening disease to a manageable chronic condition. However, modern therapies have been less successful in managing HIV in Africa, where there is more viral heterogeneity and access to sequencing is much less available. In this work, we present a novel mixture based approach that uses a deep information bottleneck to transfer patterns learned from European HIV cohorts where genomic data is readily available to African patients where no such data is available. We demonstrate its utility for optimising treatments for the first time in a set of HIV patients in Africa, and note how this approach may be applicable to many other scenarios where a variable is measured in some population but is missing from the target population.