Supervised and unsupervised deep learning-based approaches for studying DNA replication spatiotemporal dynamics

Abstract

In eukaryotic cells, DNA replication is organised both spatially and temporally, as evidenced by the stage-specific spatial distribution of replication foci in the nucleus. Despite the genetic association of aberrant DNA replication with numerous human diseases, the labour-intensive methods employed to study DNA replication have hindered large-scale analyses of its roles in pathological processes. In this study, we first demonstrate that a convolutional neural network trained to classify S-phase stages based on DAPI and EdU patterns could identify altered replication dynamics in Rif1-deficient mouse embryonic stem cells (mESCs), revealing a skewed distribution across the various S-phase stages. Given the possible practical limitations associated with a supervised framework, we proceed to show that the abnormal replication profile of Rif1-deficient mESCs could further be detected by an unsupervised approach (based on self-supervised representation learning), which could additionally reconstruct progression through S-phase. Finally, we extend our approach to a well-characterised cellular model of inducible deregulated origin firing, involving cyclin E overexpression. Through parallel EdU- and PCNA-based analyses, we demonstrate the potential applicability of our method to patient samples, offering a means to identify the contribution of deregulated DNA replication to a plethora of pathogenic processes.