Speaker: Wolfgang M. A. Pernice, Ph.D., Columbia University, Departement of Neurology

Title: Deep learning for the rapid detection of disease-associated phenotypes in primary cell-culture models through high-content morphological profiling

Abstract:

Despite abundant sequencing data, our understanding of the genetic landscape of neurodegenerative diseases remains woefully incomplete. While gene discovery studies typically result in numerous candidates, traditional gene-by-gene strategies for experimental validation quickly become infeasible at scale. We hence sought to develop an unbiased, high-content approach to a) rapidly screen patient derived cells for distinguishing phenotypes that allow the experimental validation of candidate genetic variants and that b) allows inference of sufficient biological context to enable targeted follow-up studies for genes with unknown function. We here apply supervised machine learning methods as an unbiased, high-content approach to mine high-resolution, multiplexed fluorescent microscopy images of individual cells in primary culture, for distinguishing (testable) morphological phenotypes, that allow their classification according to genotype. In a prototype study we demonstrate that this approach can a) achieve super-human classification accuracy, b) allows for rapid phenotypic evaluation of cells with mutations unbeknownst to the network, and we pioneer its application to primary cells derived from patients with neurodegenerative conditions.