Search

Due to the powerful feature expression ability of deep learning and its end-to-end nonlinear mapping relationship, deep-learning-based methods have become the mainstream method for hyperspectral image (HSI) classification tasks. However, the accuracy of deep learning methods greatly depends on the use of a large number of labeled samples to train the model. Also, HSIs have few labeled samples and unbalanced categories, which make the depth model prone to overfittingand seriously affect the classification accuracy. Therefore, how to alleviate the overfitting phenomenon caused by small samples in the classification problem based on deep learning is still a problem that needs to be solved. Considering that it is relatively easier to obtain a large number of unlabeled samples in the field of remote sensing, making full use of the unsupervised information learned from unlabeled data can regularize the supervised classification model, which can effectively alleviate the overfitting phenomenon caused by the small samples problem. In the supervised training process, unsupervised information from the overall distribution of the sample is introduced to guide the regularization of the model, so as to realize the effective classification of the data in the case of a small number of labeled samples. Experimental results demonstrate the effectiveness of the proposed method in terms of HSI classification with few training samples.

Abstract Background Apoptosis-inducing factor (AIF), as a mitochondrial flavoprotein, plays a fundamental role in mitochondrial bioenergetics that is critical for cell survival and also mediates caspase-independent cell death once it is released from mitochondria and translocated to the nucleus under ischemic stroke or neurodegenerative diseases. Although alternative splicing regulation of AIF has been implicated, it remains unknown which AIF splicing isoform will be induced under pathological conditions and how it impacts mitochondrial functions and neurodegeneration in adult brain. Methods AIF splicing induction in brain was determined by multiple approaches including 5′ RACE, Sanger sequencing, splicing-specific PCR assay and bottom-up proteomic analysis. The role of AIF splicing in mitochondria and neurodegeneration was determined by its biochemical properties, cell death analysis, morphological and functional alterations and animal behavior. Three animal models, including loss-of-function harlequin model, gain-of-function AIF3 knockin model and conditional inducible AIF splicing model established using either Cre-loxp recombination or CRISPR/Cas9 techniques, were applied to explore underlying mechanisms of AIF splicing-induced neurodegeneration. Results We identified a nature splicing AIF isoform lacking exons 2 and 3 named as AIF3. AIF3 was undetectable under physiological conditions but its expression was increased in mouse and human postmortem brain after stroke. AIF3 splicing in mouse brain caused enlarged ventricles and severe neurodegeneration in the forebrain regions. These AIF3 splicing mice died 2–4 months after birth. AIF3 splicing-triggered neurodegeneration involves both mitochondrial dysfunction and AIF3 nuclear translocation. We showed that AIF3 inhibited NADH oxidase activity, ATP production, oxygen consumption, and mitochondrial biogenesis. In addition, expression of AIF3 significantly increased chromatin condensation and nuclear shrinkage leading to neuronal cell death. However, loss-of-AIF alone in harlequin or gain-of-AIF3 alone in AIF3 knockin mice did not cause robust neurodegeneration as that observed in AIF3 splicing mice. Conclusions We identified AIF3 as a disease-inducible isoform and established AIF3 splicing mouse model. The molecular mechanism underlying AIF3 splicing-induced neurodegeneration involves mitochondrial dysfunction and AIF3 nuclear translocation resulting from the synergistic effect of loss-of-AIF and gain-of-AIF3. Our study provides a valuable tool to understand the role of AIF3 splicing in brain and a potential therapeutic target to prevent/delay the progress of neurodegenerative diseases.

Galanin (GAL) plays key role in many pathophysiological processes, but its role in ischemic stroke remains unclear. Here, the models of 1 h middle cerebral artery occlusion (MCAO)/1-7 d reperfusion (R)-induced ischemic stroke and in vitro cell ischemia of 1 h oxygen-glucose deprivation (OGD)/24 h reoxygenation in primary cultured cortical neurons were used to explore GAL's effects and its underlying mechanisms. The results showed significant increases of GAL protein levels in the peri-infarct region (P) and infarct core (I) within 48 h R of MCAO mice (p<0.001). The RT-qPCR results also demonstrated significant increases of GAL mRNA during 24-48 h R (p<0.001), and GAL receptors GalR1-2 (but not 3) mRNA levels in the P region at 24 h R of MCAO mice (p<0.001). Furthermore, the significant decrease of infarct volume (p<0.05) and improved neurological outcome (p<0.001-0.05) were observed in MCAO mice following 1 h pre- or 6 h post-treatment of GAL during 1-7 d reperfusion. GalR1 was confirmed as the receptor responsible for GAL-induced neuroprotection by using GalR2/3 agonist AR-M1896 and Lentivirus-based RNAi knockdown of GalR1. GAL treatment inhibited Caspase-3 activation through the upstream initiators Capsases-8/-12 (not Caspase-9) in both P region and OGD-treated cortical neurons. Meanwhile, GAL's neuroprotective effect was not observed in cortical neurons from conventional protein kinase C (cPKC) γ knockout mice. These results suggested that exogenous GAL protects the brain from ischemic injury by inhibiting Capsase-8/12-initiated apoptosis, possibly mediated by GalR1 via the cPKCγ signaling pathway. [ABSTRACT FROM AUTHOR]