Characterizing chromatin folding coordinate and landscape with deep learning
WJ Xie and YF Qi and B Zhang, PLOS COMPUTATIONAL BIOLOGY, 16, e1008262 (2020).
Author summary Chromatin folding, the dynamical process during which chromatin establishes its three-dimensional organization for proper function, is of critical importance. However, it is difficult to visualize and characterize due to challenges associated with live-cell imaging at high temporal and spatial resolution. Here, using a combination of deep learning and statistical mechanical theory, we demonstrate that great insight can be gained into the folding process by analyzing snapshots of chromatin structures taken across a population of cells. Though these static structures are not connected in time, prior research on chemical reactions suggests that fluctuation within the conformational ensemble provides valuable information for uncovering the reaction mechanism. Our analysis reconciles the seemingly contradictory results from different experimental techniques and supports the presence of multiple factors in organizing the chromatin. As single-cell experimental data are becoming routine, the approaches presented here could help with their interpretation to provide more insight into chromatin folding. Genome organization is critical for setting up the spatial environment of gene transcription, and substantial progress has been made towards its high-resolution characterization. The underlying molecular mechanism for its establishment is much less understood. We applied a deep-learning approach, variational autoencoder (VAE), to analyze the fluctuation and heterogeneity of chromatin structures revealed by single-cell imaging and to identify a reaction coordinate for chromatin folding. This coordinate connects the seemingly random structures observed in individual cohesin-depleted cells as intermediate states along a folding pathway that leads to the formation of topologically associating domains (TAD). We showed that folding into wild-type-like structures remain energetically favorable in cohesin- depleted cells, potentially as a result of the phase separation between the two chromatin segments with active and repressive histone marks. The energetic stabilization, however, is not strong enough to overcome the entropic penalty, leading to the formation of only partially folded structures and the disappearance of TADs from contact maps upon averaging. Our study suggests that machine learning techniques, when combined with rigorous statistical mechanical analysis, are powerful tools for analyzing structural ensembles of chromatin.
Return to Publications page