Identifying meaningful pairwise relationships between entities from very high-dimensional data has become a central task in data science. I will describe the approach usually called "direct coupling analysis" (DCA) which has been successfully used to infer residue-residue contacts in proteins and recently also for in silico protein structure prediction (for a recent state-of-the art, see Ovchinnikov et al, Science 355: 294--298 (2017)).
I will give an interpretation of the DCA as an analysis of epistasis, and discuss an application to bacterial genomics.
The presentation is based on MJ Skwark et al. PLoS Genetics vol 13 (2), e1006508 (2017)
