Direct training high-performance deep spiking neural networks: a review of theories and methods.

Spiking neural networks (SNNs) offer a promising energy-efficient alternative to artificial neural networks (ANNs), in virtue of their high biological plausibility, rich spatial-temporal dynamics, and event-driven computation. The direct training algorithms based on the surrogate gradient method provide sufficient flexibility to design novel SNN architectures and explore the spatial-temporal dynamics of SNNs. According to previous studies, the performance of models is highly dependent on their sizes. Recently, direct training deep SNNs have achieved great progress on both neuromorphic datasets and large-scale static datasets. Notably, transformer-based SNNs show comparable performance with their ANN counterparts. In this paper, we provide a new perspective to summarize the theories and methods for training deep SNNs with high performance in a systematic and comprehensive way, including theory fundamentals, spiking neuron models, advanced SNN models and residual architectures, software frameworks and neuromorphic hardware, applications, and future trends.

SGLFormer: Spiking Global-Local-Fusion Transformer with high performance.

Introduction: Spiking Neural Networks (SNNs), inspired by brain science, offer low energy consumption and high biological plausibility with their event-driven nature. However, the current SNNs are still suffering from insufficient performance. Methods: Recognizing the brain's adeptness at information processing for various scenarios with complex neuronal connections within and across regions, as well as specialized neuronal architectures for specific functions, we propose a Spiking Global-Local-Fusion Transformer (SGLFormer), that significantly improves the performance of SNNs. This novel architecture enables efficient information processing on both global and local scales, by integrating transformer and convolution structures in SNNs. In addition, we uncover the problem of inaccurate gradient backpropagation caused by Maxpooling in SNNs and address it by developing a new Maxpooling module. Furthermore, we adopt spatio-temporal block (STB) in the classification head instead of global average pooling, facilitating the aggregation of spatial and temporal features. Results: SGLFormer demonstrates its superior performance on static datasets such as CIFAR10/CIFAR100, and ImageNet, as well as dynamic vision sensor (DVS) datasets including CIFAR10-DVS and DVS128-Gesture. Notably, on ImageNet, SGLFormer achieves a top-1 accuracy of 83.73% with 64 M parameters, outperforming the current SOTA directly trained SNNs by a margin of 6.66%. Discussion: With its high performance, SGLFormer can support more computer vision tasks in the future. The codes for this study can be found in https://github.com/ZhangHanN1/SGLFormer.

DCAF1 regulates Treg senescence via the ROS axis during immunological aging.

As a hallmark of immunological aging, low-grade, chronic inflammation with accumulation of effector memory T cells contributes to increased susceptibility to many aging-related diseases. While the proinflammatory state of aged T cells indicates a dysregulation of immune homeostasis, whether and how aging drives regulatory T cell (Treg) aging and alters Treg function are not fully understood owing to a lack of specific aging markers. Here, by a combination of cellular, molecular, and bioinformatic approaches, we discovered that Tregs senesce more severely than conventional T (Tconv) cells during aging. We found that Tregs from aged mice were less efficient than young Tregs in suppressing Tconv cell function in an inflammatory bowel disease model and in preventing Tconv cell aging in an irradiation-induced aging model. Furthermore, we revealed that DDB1- and CUL4-associated factor 1 (DCAF1) was downregulated in aged Tregs and was critical to restrain Treg aging via reactive oxygen species (ROS) regulated by glutathione-S-transferase P (GSTP1). Importantly, interfering with GSTP1 and ROS pathways reinvigorated the proliferation and function of aged Tregs. Therefore, our studies uncover an important role of the DCAF1/GSTP1/ROS axis in Treg senescence, which leads to uncontrolled inflammation and immunological aging.

PHB Associates with the HIRA Complex to Control an Epigenetic-Metabolic Circuit in Human ESCs.

The chromatin landscape and cellular metabolism both contribute to cell fate determination, but their interplay remains poorly understood. Using genome-wide siRNA screening, we have identified prohibitin (PHB) as an essential factor in self-renewal of human embryonic stem cells (hESCs). Mechanistically, PHB forms protein complexes with HIRA, a histone H3.3 chaperone, and stabilizes the protein levels of HIRA complex components. Like PHB, HIRA is required for hESC self-renewal. PHB and HIRA act together to control global deposition of histone H3.3 and gene expression in hESCs. Of particular note, PHB and HIRA regulate the chromatin architecture at the promoters of isocitrate dehydrogenase genes to promote transcription and, thus, production of α-ketoglutarate, a key metabolite in the regulation of ESC fate. Our study shows that PHB has an unexpected nuclear role in hESCs that is required for self-renewal and that it acts with HIRA in chromatin organization to link epigenetic organization to a metabolic circuit.

Comprehensive profiling reveals mechanisms of SOX2-mediated cell fate specification in human ESCs and NPCs.

SOX2 is a key regulator of multiple types of stem cells, especially embryonic stem cells (ESCs) and neural progenitor cells (NPCs). Understanding the mechanism underlying the function of SOX2 is of great importance for realizing the full potential of ESCs and NPCs. Here, through genome-wide comparative studies, we show that SOX2 executes its distinct functions in human ESCs (hESCs) and hESC-derived NPCs (hNPCs) through cell type- and stage-dependent transcription programs. Importantly, SOX2 suppresses non-neural lineages in hESCs and regulates neurogenesis from hNPCs by inhibiting canonical Wnt signaling. In hESCs, SOX2 achieves such inhibition by direct transcriptional regulation of important Wnt signaling modulators, WLS and SFRP2. Moreover, SOX2 ensures pluripotent epigenetic landscapes via interacting with histone variant H2A.Z and recruiting polycomb repressor complex 2 to poise developmental genes in hESCs. Together, our results advance our understanding of the mechanism by which cell type-specific transcription factors control lineage-specific gene expression programs and specify cell fate.

Stk40 represses adipogenesis through translational control of CCAAT/enhancer-binding proteins.

A better understanding of molecular regulation in adipogenesis might help the development of efficient strategies to cope with obesity-related diseases. Here, we report that CCAAT/enhancer-binding protein (C/EBP) ß and C/EBPδ, two crucial pro-adipogenic transcription factors, are controlled at a translational level by serine/threonine kinase 40 (Stk40). Genetic knockout (KO) or knockdown (KD) of Stk40 leads to increased protein levels of C/EBP proteins and adipocyte differentiation in mouse embryonic fibroblasts (MEFs), fetal liver stromal cells, and mesenchymal stem cells (MSCs). In contrast, overexpression of Stk40 abolishes the enhanced C/EBP protein translation and adipogenesis observed in Stk40-KO and -KD cells. Functionally, knockdown of C/EBPß eliminates the enhanced adipogenic differentiation in Stk40-KO and -KD cells substantially. Mechanistically, deletion of Stk40 enhances phosphorylation of eIF4E-binding protein 1, leading to increased eIF4E-dependent translation of C/EBPß and C/EBPδ. Knockdown of eIF4E in MSCs decreases translation of C/EBP proteins. Moreover, Stk40-KO fetal livers display an increased adipogenic program and aberrant lipid and steroid metabolism. Collectively, our study uncovers a new repressor of C/EBP protein translation as well as adipogenesis and provides new insights into the molecular mechanism underpinning the adipogenic program.

Nucleolin maintains embryonic stem cell self-renewal by suppression of p53 protein-dependent pathway.

Embryonic stem cells (ESCs) can undergo unlimited self-renewal and retain pluripotent developmental potential. The unique characteristics of ESCs, including a distinct transcriptional network, a poised epigenetic state, and a specific cell cycle profile, distinguish them from somatic cells. However, the molecular mechanisms underlying these special properties of ESCs are not fully understood. Here, we report that nucleolin, a nucleolar protein highly expressed in undifferentiated ESCs, plays an essential role for the maintenance of ESC self-renewal. When nucleolin is knocked down by specific short hairpin RNA (shRNA), ESCs display dramatically reduced cell proliferation rate, increased cell apoptosis, and G(1) phase accumulation. Down-regulation of nucleolin also leads to evident ESC differentiation as well as decreased self-renewal ability. Interestingly, expression of pluripotency markers (Oct4 and Nanog) is unaltered in these differentiated cells. Mechanistically, depletion of nucleolin up-regulates the p53 protein level and activates the p53-dependent pathway, at least in part, via increasing p53 protein stability. Silencing of p53 rescues G(1) phase accumulation and apoptosis caused by nucleolin deficiency entirely, although it partially blocks abnormal differentiation in nucleolin-depleted ESCs. It is noteworthy that knocking down nucleolin in NIH3T3 cells affected cell survival and proliferation in a much milder way, despite the comparable silencing efficiency obtained in ESCs and NIH3T3 cells. Collectively, our data demonstrate that nucleolin is a critical regulator of ESC self-renewal and that suppression of the p53-dependent pathway is the major molecular mechanism underlying functions of nucleolin in ESCs.