A University of Pennsylvania-led team narrows in on more than a dozen proposed effector genes at COVID-19-related genetic loci with the help of a "variant-to-gene mapping" method applied to human immune cell types implicated in disease susceptibility or severity. Using a combination of ATAC-seq, high-resolution chromatin conformation capture focused on promoter regions, and other approaches, the researchers tracked down potential regulatory sequences at COVID-19 severity-associated loci and linked these regions to 16 suspected effector genes from pathways involved in everything from inflammation and interferon response to viral replication. "Together," they say, "our results suggest that genes central to viral genome sensing, host control of viral replication, the interferon response, and immune inflammation are likely under genetic control by common variants associated with COVID-19 disease risk."
A team at the University of Toronto and the Vector Institute presents a Virtual ChIP-seq method designed for translating transcriptome data into predicted transcription factor binding patterns — an approach they use to predict binding for nearly three dozen chromatin factors across 33 tissue types included in the Roadmap Epigenomics project. "Although Virtual ChIP-seq uses direct evidence of chromatin factor binding at each genomic region as one of the input features, it is able to correctly predict new peaks which do not exist in training cell types," the authors note, adding that datasets produced with the approach so far "should also accelerate the development of future machine4 learning methods by many groups."
Researchers at the Georgia Institute of Technology and the University of Tokyo introduce a deep learning framework for bringing together single-cell RNA sequence and single-cell ATAC-seq data. The scalable approach — known as "single-cell deep learning model for ATAC-seq and RNA-seq trajectory integration," or scDART — "integrates scRNA-seq and scATAC-seq data and learns cross-modalities relationships simultaneously," the team writes. The authors went on to apply scDART to simulated and real datasets, including a dataset with matched gene expression and chromatin accessibility profiles from the same cells and datasets with independent scRNA-seq and scATAC-seq data. "In the era of single-cell multi-omics, the goal of data integration should be not only removing batch effects and compiling larger datasets, but also learning the relationship between different data modalities," the authors suggest.