Charles K. Fisher

The paper uses a Conditional Restricted Boltzmann Machine to generate digital twins of MS patients, enabling subject-level analysis of disease progression.

The paper evaluates representation learning techniques for transcriptomics-based phenotype prediction using public data and finds that l2-regularized regression on centered log-ratio transformed abundances performs best.

The paper introduces a Python library, genemunge, to simplify handling transcriptomics data by managing gene identifiers, GO information, expression levels, batch correction, and gene set selection.