Development of digital Hi-C assay
Abstracts Enhancers are genomic elements and contain all necessary cis-regulatory contexts. Such enhancers are convened to the appropriate promoter of target genes for gene regulations even though the enhancers and the promoters are apart a few mega-base pairs away from each other. In addition to physical distance, nucleotide mutations in enhancers influence a partial group of the target genes. Those make it more complicated to reveal the paired relationship between enhancer and promoter of target genes. Recently, advanced computational approaches are employed to predict such interactions. One approach requires a large number of different high-throughput datasets to predict such interactions; however, in practical aspects, all datasets for tissues and conditions of interest are not available. Whereas the alternative approach requires only genome sequences for particular predictions, their predictions are insufficient for practical applications. We address those issues by developing the digital Hi-C assay with a transformer-algorithm basis. This assay allows us to create models from simple/small/limited sequence-based datasets only. We apply the trained models to be able to identify long-distance interactions of genomic loci and three-dimensional (3D) genomic architectures in any other tissue/cell datasets; additionally, we demonstrated the predictions of genomic contexts by analysing the prediction patterns around the target locus in the three following genomic-context problems: enhancer-promoter interactions (i.e., promoter-capture Hi-C), the CTCF-enriched regions, and TAD-boundary regions. Because our approach adopted a sequence-based approach, we can predict the long-distance interactions of genomic loci by using the genomic sequences of the users interest (e.g., input sequences from high-throughput assay datasets such as ATAC-seq and ChIP-seq assays). Consequently, we provide an opportunity to predict interactions of genomic loci from a minimum dataset.