Sequence-based modeling of genome 3D architecture from kilobase to chromosome-scale

Abstract

The structural organization of the genome plays an important role in multiple aspects of genome function. Understanding how genomic sequence influences 3D organization can help elucidate their roles in various processes in healthy and disease states. However, the sequence determinants of genome structure across multiple spatial scales are still not well understood. To learn the complex sequence dependencies of multiscale genome architecture, here we developed a sequence-based deep learning approach, Orca, that predicts genome 3D architecture from kilobase to whole-chromosome scale, covering structures including chromatin compartments and topologically associating domains. Orca also makes both intrachromosomal and interchromosomal predictions and captures the sequence dependencies of diverse types of interactions, from CTCF-mediated to enhancer-promoter interactions and Polycomb-mediated interactions. Orca enables the interpretation of the effects of any structural variant at any size on multiscale genome organization and provides an in silico model to help study the sequence-dependent mechanistic basis of genome architecture. We show that the models accurately recapitulate effects of experimentally studied structural variants at varying sizes (300bp-80Mb) using only sequence. Furthermore, these sequence models enable in silico virtual screen assays to probe the sequence-basis of genome 3D organization at different scales. At the submegabase scale, the models predicted specific transcription factor motifs underlying cell-type-specific genome interactions. At the compartment scale, based on virtual screens of sequence activities, we propose a new model for the sequence basis of chromatin compartments: sequences at active transcription start sites are primarily responsible for establishing the expression-active compartment A, while the inactive compartment B typically requires extended stretches of AT-rich sequences (at least 6-12kb) and can form 'passively' without depending on any particular sequence pattern. Orca thus effectively provides an 'in silico genome observatory' to predict variant effects on genome structure and probe the sequence-based mechanisms of genome organization.

Regulation associated modules reflect 3D genome modularity associated with chromatin activity

“…To systematically examine the impact of deleting cRAM boundaries to the chromatin structure, we resorted to computational predictions using a deep learning model ORCA( 38 ) (https://github.com/jzhoulab/orca) as it is prohibitive to perform hundreds of Hi-C experiments with sufficient resolution. We took the ORCA model pre-trained on the high resolution Hi-C and Micro-C data in H1-hESC and HFF cell lines to predict 3D chromatin architecture from kilobase to whole-chromosome scale using DNA sequences.…”

Reconstructing the 3D genome organization of Neanderthals reveals that chromatin folding shaped phenotypic and sequence divergence

“…Finally, we develop a method to robustly identify 3D divergent windows between populations. With the recent growth of 3D genome in silico predictors [81][82][83][84], we anticipate that our work can provide a foundation for both hypothesis generation and prioritization of experimental resources.…”

“…3D genome folding is facilitated by a complex interplay of CTCF binding with cohesin and other architectural factors [50,62,79,80]. Recent deep learning methods have been developed that learn the sequence "grammar" underlying 3d genome folding patterns [81][82][83][84]. We hypothesized that these deep learning methods would allow us to infer genome-wide 3D chromatin contact maps of Neanderthals and Denisovans.…”

Machine learning reveals the diversity of human 3D chromatin contact patterns

“…This has limited the extent to which chromatin contact diversity has been studied across human populations. However, recent advances in machine learning methods have allowed for the prediction of 3D genome chromatin contact maps from DNA sequences (Fudenberg et al, 2020; Schwessinger et al, 2020; Zhou, 2021). These methods predict 3D chromatin contact based solely on sequence information, offering a promising approach to computationally study 3D genome diversity.…”