Chromatin Folding Coordinate and Landscape Unraveled by Deep Learning Analysis of Single-Cell Imaging Data

Abstract

Chromatin spatial organization is critical in various DNA-templated processes, and substantial progress has been made towards its characterization. However, the underlying molecular mechanism for its establishment is much less understood. Here we combine the energy landscape theory with deep learning techniques to investigate the mechanism of genome folding. Specifically, we analyze single-cell imaging data with the variational autoencoder (VAE). The one-dimensional folding coordinate obtained from VAE provides a new perspective on the heterogeneity of wild-type chromatin configurations and connects the seemingly random structures in cohesin-depleted cells as intermediate states along the folding pathway. We also found folded structures in cohesin-depleted cells that have the same domain boundary as those in wild-type cells. The folding landscape derived from VAE suggests that folded structures in cohesin-depleted cells remain energetically favorable which might originate from a phase separation mechanism. Our study, therefore, opens up new directions for mechanistic investigation of chromatin folding.