Recent advances in long-range Hi-C contact mapping have revealed the importance of the 3D structure of chromosomes in gene expression. model also reveals the existence of several negative drivers that counteract the presence of domain borders including P300, RXRA, ELK1 and BCL11A. Author Summary Chromosomal DNA is tightly packed up in 3D such that around 2 meters of this long molecule fits into the microscopic nucleus of every cell. The genome packing is not random, but instead structured in 3D domains that are essential to numerous key processes in the cell, such as for the regulation of gene expression or for the replication of DNA. A current challenge is to identify the key molecular drivers of this higher-order chromosome organization. Here we propose a novel computational integrative approach to identify proteins and DNA elements that positively or negatively influence the establishment or maintenance of 3D domains. Analysis of data at very high resolution suggests that among architectural proteins, BEAF-32 and CP190 are the main positive drivers of 3D domains. In humans, our results highlight the roles of CTCF, cohesin, ZNF143 and Polycomb group proteins as positive drivers of 3D domains, in contrast to P300, RXRA, BCL11A and ELK1 that act as negative drivers. Introduction High-throughput chromatin conformation capture (Hi-C) has emerged over the past years as an efficient approach to map long-range chromatin contacts [1C3]. This technique has allowed the study of the 3D architecture Givinostat of chromosomes at an unprecedented resolution for many genomes and cell types [4C7]. Multiple hierarchical levels of genome organization have been revealed: compartments A/B [1], sub-compartments [8], topologically associating domains (TADs) [4, 5] and sub-TADs [7]. Among those domains, TADs represent a pervasive structural feature of the genome organization. TADs are stable across different cell types and highly conserved across species. A current challenge is to identify the molecular drivers of topological arrangements of higher-order chromatin organization. There is a growing body of evidence that insulator binding proteins (IBPs) such as CTCF, and cofactors such as cohesin, act as mediators of long-range chromatin contacts [5, 6, 9C11]. In human, depletion of cohesin predominantly reduces interactions within TADs, whereas depletion of CTCF not only decreases intradomain contacts but also increases interdomain contacts [12]. The densest Hi-C mapping in human has recently revealed that loops that demarcate domains are often marked by asymmetric CTCF motifs where cohesin is recruited [8]. In and human Hi-C data allowing to probe TAD borders depending on multiple proteins and functional elements. Using both simulated and real data, we show that our model outperforms enrichment test and non-parametric models such as random forests for the identification of known and suspected architectural proteins. In addition, the proposed method identifies genomic features that positively or negatively impact TAD borders with a very high resolution of 1 kb. Results The model The proposed multiple logistic regression models the influences of genomic features on 3D domain borders: genomic features Givinostat such as DNA-binding proteins and is a variable that indicates if the genomic bin belongs to a border (= 1) or not (= 0). The set = {> 0 Givinostat and the parameter associated with protein B > 0. In other words, both proteins A and B are enriched at 3D domain borders. Multiple logistic regression will instead estimate that parameters > 0 and = 0. This means that protein A positively influences 3D domain borders, while protein B does not. This is because multiple logistic regression can discard spurious associations (here between protein B and 3D domain borders). One would argue that enrichment test can also be used to discard the Givinostat spurious association if the enrichment of protein B when Rabbit Polyclonal to RHOB protein A is absent is tested instead. However such conditional enrichment test becomes intractable when more than 3 proteins colocalize to domain borders, whereas.