Project Summary/Abstract Chromatin 3D structure plays an important role in fundamental genomics processes. A variety of experimental methods, such as Hi-C, GAM, SPRITE, and HiChIP, have been developed to characterize chromatin architecture and DNA-DNA interactions genome-wide. Meanwhile, other types of data, such as gene expression and chromatin accessibility profiles, have also been used to refine our understanding of gene regulation and chromatin structure. However, computational tools that can jointly analyze Hi-C and other types of data are still lacking, hindering the process of comprehensively understanding the relationship between genome structure and function. Moreover, the heterogeneous, large-scale, noisy and high-dimensional nature of these data presents computational challenges for effectively integrating Hi-C data with other types of data. Here, we propose to develop a series of machine learning models that integrate contact matrices with RNA-seq, genome sequence, and ATAC-seq data to advance chromatin structure analysis. First, to identify the dynamic in- terplay between cell-type-specific gene expression and chromatin structure, we will extend our Sagittarius model, which obtains state-of-the-art results in modeling RNA-seq time series, to analyze Hi-C time-course data by ex- plicitly modeling the time dimension. This new model will enable spatio-temporal analyses of chromatin structures for differentiation, development, and disease progression. Second, genome sequence has been successfully used to predict 3D genome folding but has not been fully exploited for resolution enhancement. We will develop a graph-based framework to co-embed genome sequences and low-coverage contact matrices for resolution en- hancement. The imputed high-resolution data will enable biologists to identify 3D chromatin features that can only be discovered at high resolution, such as punctate loops and sub-domains. Third, the view of chromatin architecture provided by a contact matrix has not been fully integrated with the linear, high-resolution picture of local chromatin architecture provided by ATAC-seq data. We will develop a translation model between Hi-C and ATAC-seq, which will be used to analyze cell types or species that only have one of these two modalities. This translation model will provide a consolidated view of 3D chromatin architecture and further advance downstream analyses of regulatory processes, such as promoter-enhancer interactions, replication timing, gene expression, and mRNA splicing. All of the software produced by this project will be open source, and all of the imputed data and pre-trained models will be made publicly available, providing a valuable resource for users interested in understanding chro- matin 3D architecture and its relationship to gene expression and other functional cellular processes.