Genome-wide mapping of chromatin interactions at high resolution remains experimentally and computationally challenging. Here we used a low-input ''easy Hi-C'' protocol to map the 3D genome architecture in human neurogenesis and brain tissues and also demonstrated that a rigorous Hi-C bias-correction pipeline (HiCorr) can significantly improve the sensitivity and robustness of Hi-C loop identification at sub-TAD level, especially the enhancer-promoter (E-P) interactions. We used HiCorr to compare the high-resolution maps of chromatin interactions from 10 tissue or cell types with a focus on neurogenesis and brain tissues. We found that dynamic chromatin loops are better hallmarks for cellular differentiation than compartment switching. HiCorr allowed direct observation of cell-type-and differentiation-specific E-P aggregates spanning large neighborhoods, suggesting a mechanism that stabilizes enhancer contacts during development. Interestingly, we concluded that Hi-C loop outperforms eQTL in explaining neurological GWAS results, revealing a unique value of high-resolution 3D genome maps in elucidating the disease etiology.

Large populations worldwide have been deprived from nature experiences due to mass quarantines and lockdowns during the COVID-19 pandemic, and face a looming mental health crisis. Virtual reality offers a safe and practical solution to increase nature exposure. This research examined the effects of virtual nature using a within-subject design with young adults (Study 1) and senior citizens (Study 2). Results from the young adult sample showed that walking in a virtual forest reduced negative affect due to enhanced nature connectedness, and reduced stress measured by heart rate. Consistently, the senior citizen sample reported improved positive affect due to enhanced nature connectedness after the virtual nature walk. Our findings unveil the underlying mechanism of how virtual nature may improve psychological well-being and demonstrated how virtual nature can be used as an intervention to promote mental health.

The in vitro differentiation of insulin-producing β-like cells can model aspects of human pancreatic development. Here we generate 95,308 single cell transcriptomes and reconstruct a lineage tree of the entire differentiation process from hESCs to β-like cells to study temporally regulated genes during differentiation. We identify so-called ‘switch genes’ at the branch point of endocrine/non-endocrine cell fate choice, revealing insights into the mechanisms of differentiation promoting reagents, such as NOTCH and ROCKII inhibitors, and providing improved differentiation protocols. Over 20% of all detectable genes are activated multiple times during differentiation, even though their enhancer activation is usually unimodal, indicating extensive gene reuse driven by different enhancers. We also identify a stage-specific enhancer in the TCF7L2 diabetes GWAS locus that drives a transient wave of gene expression in pancreatic progenitors. Finally, we develop a web app to visualize gene expression on the lineage tree, providing a comprehensive single cell data resource for researchers studying islet biology and diabetes.

In the context of electroencephalogram (EEG)-based driver drowsiness recognition, it is still challenging to design a calibration-free system, since EEG signals vary significantly among different subjects and recording sessions. Many efforts have been made to use deep learning methods for mental state recognition from EEG signals. However, existing work mostly treats deep learning models as black-box classifiers, while what have been learned by the models and to which extent they are affected by the noise in EEG data are still underexplored. In this paper, we develop a novel convolutional neural network combined with an interpretation technique that allows sample-wise analysis of important features for classification. The network has a compact structure and takes advantage of separable convolutions to process the EEG signals in a spatial-temporal sequence. Results show that the model achieves an average accuracy of 78.35% on 11 subjects for leave-one-out cross-subject drowsiness recognition, which is higher than the conventional baseline methods of 53.40%-72.68% and state-of-the-art deep learning methods of 71.75%-75.19%. Interpretation results indicate the model has learned to recognize biologically meaningful features from EEG signals, e.g., Alpha spindles, as strong indicators of drowsiness across different subjects.In addition, we also explore reasons behind some wrongly classified samples with the interpretation technique and discuss potential ways to improve the recognition accuracy. Our work illustrates a promising direction on using interpretable deep learning models to discover meaningful patterns related to different mental states from complex EEG signals.