Search

This tutorial paper surveys provably optimal alternatives to end-to-end backpropagation (E2EBP) -- the de facto standard for training deep architectures. Modular training refers to strictly local training without both the forward and the backward pass, i.e., dividing a deep architecture into several nonoverlapping modules and training them separately without any end-to-end operation. Between the fully global E2EBP and the strictly local modular training, there are weakly modular hybrids performing training without the backward pass only. These alternatives can match or surpass the performance of E2EBP on challenging datasets such as ImageNet, and are gaining increasing attention primarily because they offer practical advantages over E2EBP, which will be enumerated herein. In particular, they allow for greater modularity and transparency in deep learning workflows, aligning deep learning with the mainstream computer science engineering that heavily exploits modularization for scalability. Modular training has also revealed novel insights about learning and has further implications on other important research domains. Specifically, it induces natural and effective solutions to some important practical problems such as data efficiency and transferability estimation., Comment: Accepted by IEEE Computational Intelligence Magazine

Background Mastitis is one of the most severe diseases in humans and animals, especially on dairy farms. Mounting evidence indicates that gastrointestinal dysbiosis caused by induction of subacute ruminal acidosis (SARA) by high-grain diet consumption and low in dietary fiber is associated with mastitis initiation and development, however, the underlying mechanism remains unknown. Results In the present study, we found that cows with SARA-associated mastitis have altered metabolic profiles in the rumen, with increased sialic acids level in particular. Consumption of sialic acid (SA) in antibiotic-treated mice, but not healthy mice, induced marked mastitis. SA treatment of antibiotic-treated mice also induced mucosal and systemic inflammatory responses, as evidenced by increased colon and liver injuries and several inflammatory markers. In addition, gut dysbiosis caused by antibiotic impaired gut barrier integrity, which was aggravated by SA treatment. SA potentiated serum LPS level caused by antibiotic treatment, leading to increased activation of the TLR4-NF-κB/NLRP3 pathways in the mammary gland and colon. Moreover, SA facilitated gut dysbiosis caused by antibiotic, and especially enhanced Enterobacteriaceae and Akkermansiaceae, which correlated with mastitis parameters. Fecal microbiota transplantation from SA-antibiotic-treated mice mimicked mastitis in recipient mice. In vitro experiments showed that SA prompted Escherichia coli growth and virulence gene expression, leading to higher proinflammatory cytokine production in macrophages. Targeting the inhibition of Enterobacteriaceae by sodium tungstate or treating with the commensal Lactobacillus reuteri alleviated SA-facilitated mastitis. In addition, SARA cows had distinct ruminal microbial structure by the enrichment of SA-utilizing opportunistic pathogenic Moraxellaceae and the depletion of SA-utilizing commensal Prevotellaceae. Treating mice with the specific sialidase inhibitor zanamivir reduced SA production and Moraxellaceae abundance, and improved mastitis in mice caused by ruminal microbiota transplantation from cows with SARA-associated mastitis. Conclusions This study, for the first time, indicates that SA aggravates gut dysbiosis-induced mastitis by promoting gut microbiota disturbance and is regulated by commensal bacteria, indicating the important role of the microbiota-gut-mammary axis in mastitis pathogenesis and suggesting a potential strategy for mastitis intervention based on gut metabolism regulation.

By redefining the conventional notions of layers, we present an alternative view on finitely wide, fully trainable deep neural networks as stacked linear models in feature spaces, leading to a kernel machine interpretation. Based on this construction, we then propose a provably optimal modular learning framework for classification that does not require between-module backpropagation. This modular approach brings new insights into the label requirement of deep learning: It leverages only implicit pairwise labels (weak supervision) when learning the hidden modules. When training the output module, on the other hand, it requires full supervision but achieves high label efficiency, needing as few as 10 randomly selected labeled examples (one from each class) to achieve 94.88% accuracy on CIFAR-10 using a ResNet-18 backbone. Moreover, modular training enables fully modularized deep learning workflows, which then simplify the design and implementation of pipelines and improve the maintainability and reusability of models. To showcase the advantages of such a modularized workflow, we describe a simple yet reliable method for estimating reusability of pre-trained modules as well as task transferability in a transfer learning setting. At practically no computation overhead, it precisely described the task space structure of 15 binary classification tasks from CIFAR-10.

Intestinal microbiota-mediated aryl hydrocarbon receptor (AhR) activation plays an important role in host-microbiota interactions and disease development. However, whether AhR activation mediates infection-induced inflammation in remote organs is not clear. The purpose of this study is to assess the effects and underlying mechanism of AhR activation and gut microbiota-mediated dietary tryptophan (Trp) metabolism on infection-induced inflammation using an Escherichia coli (E. coli)-induced endometritis model in mice. We found that AhR activation by 6-formylindolo (3,2-b) carbazole (Ficz), which is an AhR agonist derived from the photooxidation of Trp, alleviated E. coli-induced endometritis by repairing barrier function and inhibiting inflammatory responses, while inhibition of AhR by CH223191, which is a synthetic AhR antagonist, aggravated E. coli-induced endometritis. Gut dysbiosis damaged AhR activation and exacerbated E. coli-induced endometritis in mice, which responded to the reduced abundance of AhR ligand producers, such as

MRG domain binding protein (MRGBP) has been proposed to participate in the development of multiple tumors. However, the role of MRGBP in colorectal cancer (CRC) still remains largely unknown. Here, we found that MRGBP expression is significantly elevated in CRC, and that higher MRGBP expression correlates with poorer survival in CRC patients. Experiments in vivo and in vitro indicated that MRGBP promotes CRC cells proliferation, migration, invasion, epithelial-mesenchymal transition (EMT) and xenograft tumor growth. Mechanically, for one thing, we discovered that MRGBP suppresses DKK1 expression, thus further activating the Wnt/β-catenin pathway in CRC cells. For another, MRGBP also enhances acetylation of NF-kB/p65 pathway. Treatment with Wnt/β-catenin and NF-kB pathways inhibitors further confirmed the mediation of these two pathways in MRGBP-promoted CRC cell processes. In conclusion, these findings together suggest that MRGBP promotes CRC progression via DKK1/Wnt/β-catenin and NF-kB/p65 pathways mediated EMT, identifying MRGBP as a promising prognostic and therapeutic target for CRC.

Metastasis is a major cause of death in patients with colorectal cancer (CRC). Cysteine-rich protein 2 (CSRP2) has been recently implicated in the progression and metastasis of a variety of cancers. However, the biological functions and underlying mechanisms of CSRP2 in the regulation of CRC progression are largely unknown. Methods: Immunohistochemistry, quantitative real-time polymerase chain reaction (qPCR) and Western blotting (WB) were used to detect the expression of CSRP2 in CRC tissues and paracancerous tissues. CSRP2 function in CRC was determined by a series of functional tests in vivo and in vitro. WB and immunofluorescence were used to determine the relation between CSRP2 and epithelial-mesenchymal transition (EMT). Co-immunoprecipitation and scanning electron microscopy were used to study the molecular mechanism of CSRP2 in CRC. Results: The CSRP2 expression level in CRC tissues was lower than in adjacent normal tissues and indicated poor prognosis in CRC patients. Functionally, CSRP2 could suppress the proliferation, migration, and invasion of CRC cells in vitro and inhibit CRC tumorigenesis and metastasis in vivo. Mechanistic investigations revealed a physical interaction between CSRP2 and p130Cas. CSRP2 could inhibit the activation of Rac1 by preventing the phosphorylation of p130Cas, thus activating the Hippo signaling pathway, and simultaneously inhibiting the ERK and PAK/LIMK/cortactin signaling pathways, thereby inhibiting the EMT and metastasis of CRC. Rescue experiments showed that blocking the p130Cas and Rac1 activation could inhibit EMT induced by CSRP2 silencing. Conclusion: Our results suggest that the CSRP2/p130Cas/Rac1 axis can inhibit CRC aggressiveness and metastasis through the Hippo, ERK, and PAK signaling pathways. Therefore, CSRP2 may be a potential therapeutic target for CRC.

We propose a novel family of connectionist models based on kernel machines and consider the problem of learning layer by layer a compositional hypothesis class (i.e., a feedforward, multilayer architecture) in a supervised setting. In terms of the models, we present a principled method to “kernelize” (partly or completely) any neural network (NN). With this method, we obtain a counterpart of any given NN that is powered by kernel machines instead of neurons. In terms of learning, when learning a feedforward deep architecture in a supervised setting, one needs to train all the components simultaneously using backpropagation (BP) since there are no explicit targets for the hidden layers (Rumelhart, Hinton, & Williams, 1986 ). We consider without loss of generality the two-layer case and present a general framework that explicitly characterizes a target for the hidden layer that is optimal for minimizing the objective function of the network. This characterization then makes possible a purely greedy training scheme that learns one layer at a time, starting from the input layer. We provide instantiations of the abstract framework under certain architectures and objective functions. Based on these instantiations, we present a layer-wise training algorithm for an [Formula: see text]-layer feedforward network for classification, where [Formula: see text] can be arbitrary. This algorithm can be given an intuitive geometric interpretation that makes the learning dynamics transparent. Empirical results are provided to complement our theory. We show that the kernelized networks, trained layer-wise, compare favorably with classical kernel machines as well as other connectionist models trained by BP. We also visualize the inner workings of the greedy kernelized models to validate our claim on the transparency of the layer-wise algorithm.

Polyurethane solid–solid phase change material (PCM) with low phase change temperature is synthesized via prepolymer method. The synthetic polyurethane solid–solid PCM had lower crystallinity than PTMEG2000. The polyurethane solid–solid PCM was still in solid state during the heating process, and no liquid leakage was observed. After many phase change cycles, the polyurethane solid–solid PCM exhibited good stability of phase change cycles. Hence, the asphalt modified by polyurethane solid–solid PCM was prepared by high-speed shearing method. With increasing polyurethane solid–solid PCM proportion, the high temperature deformation resistance of modified asphalt gradually increased. At low temperature ( 3 wt%, the specific heat capacity of the modified asphalt had an evident peak ranging from 13 °C to 25 °C. In the range of the proportion, the coefficient of the heat conductivity of the modified asphalt gradually increased with increasing PCM content. During the cooling process, the modified asphalt cooled relatively slow before the phase transition of the polyurethane solid–solid PCM. With continued decrease in the temperature, the phase transition of the polyurethane solid–solid PCM occurred, and the cooling rate of the modified asphalt further decreased. With increasing polyurethane solid–solid PCM content, the temperature control ability of modified asphalt and the efficiency of temperature control gradually increased.

Bisphenol A (BPA) is a common environmental contaminant, whose exposure is associated with the progression of various kidney diseases. BPA exposure has turned out to be associated with cytotoxicity to renal tubular epithelial cells, but its underlying mechanism remains unknown. Herein, we found that BPA induced ferroptosis in kidney and renal tubular epithelial cells, as showed by increased intracellular iron accumulation, lipid peroxidation and cells death upon BPA exposure. Additionally, utilization of ferrostatin-1 and desferrioxamine, typical ferroptosis inhibitors, can fundamentally diminish cells death. Intriguingly, we discovered that autophagy inhibitor chloroquine can shield renal tubular epithelial cells from BPA-caused ferroptosis. Furthermore, we found that ferritinophagy, a phenomenon that degradation of ferritin and inducing subsequent iron overload, occurred after BPA exposure and excessive iron promoted ferroptosis through Fenton reaction. We next demonstrated that BPA activated the AMPK-mTOR-ULK1 signaling pathway. In turn, AMPK, mTOR, and ULK1 knockdown dramatically mitigated BPA-induced TCMK-1 cells death, and decreased MDA and LC3 levels, but increased FTH protein content. These results indicate that activation of the AMPK-mTOR-ULK1 signaling is involved in BPA-induced ferritinophagy. In conclusion, renal dysfunction and renal tubular epithelial damage induced by BPA are linked to ferroptosis, which depends on the activation of ferritinophagy through AMPK-mTOR-ULK1 axis.

Resistance to radiotherapy remains a severe obstacle in the treatment of high-risk colorectal cancer patients. Recent studies have indicated that proteasome activator complex subunit 3 (PSME3) participates in the development and progression of various human malignancies and is proposed to play a role in tumor radioresistance. However, the impact of PSME3 on radioresistance of colorectal cancer has been largely unknown. In the present study, the enhanced expression of PSME3 was observed in colorectal cancer cells and tissue. Upregulation of PSME3 was significantly implicated in lymph node state, lymphovascular invasion, and Dukes' stage. Furthermore, high PSME3 expression was closely linked to poorer overall and progression-free survival in patients with colorectal cancer. The study further demonstrated that the proliferative, invasive and migratory potential of colorectal cancer cells was effectively inhibited in vitro after silencing PSME3. Our results verified that knockdown of PSME3 probably triggered cell cycle arrest at the G2/M phase by downregulation of cyclinB1 and CDK1, thereby enhancing the radiosensitivity of colorectal cancer cells. These data illustrated that PSME3 is a promising biomarker predictive of colorectal cancer prognosis and silencing of PSME3 may provide with a new approach for sensitizing the radiotherapy in colorectal cancer. Impact statement It is reported that colorectal cancer (CRC) is the third most common cancer worldwide and the fourth leading cause of cancer-related death. At present, the main treatment method of colorectal cancer is surgery, supplemented by radiotherapy and chemotherapy. Among them, radiotherapy plays an important role in the treatment of locally advanced colorectal cancer, surgery, and chemotherapy. Our study found that down-regulation of PSME3 may enhance the radiosensitivity of CRC cells by triggering cell cycle arrest, which suggests that silence PSME3 may provide a new method for improving the radiosensitivity of CRC. What’more, our research also demonstrated that PSME3 may promote proliferation, invasive and migratory potential of CRC cells, which implies that PSME3 might be a biomarker of CRC for early diagnosis and treatment.

With the continuous improvement of technology in the molecular field, more and more evidence indicated that long noncoding RNAs (lncRNAs) are widely expressed in a broad spectrum of human tumors, playing an important role in the development and progression of tumors. Most studies reported that lncRNAs might serve as reliable biomarkers and effective clinical therapeutic target. Leucine-rich repeat containing 75 A-antisense RNA1 (LRRC75A-AS1) was reported to be relevant to many types of cancers and indicated to do influence on colorectal carcinoma (CRC). This research firstly examined the role of LRRC75A-AS1 in CRC and analyzed its association with the biological behaviors of CRC cells. Quantitative reverse transcription polymerase chain reaction (qRT-PCR) brought to light that LRRC75A-AS1 was remarkably expressed at low levels in CRC tissues. We also found that LRRC75A-AS1 was localized in the cytoplasm. In addition, LRRC75A-AS1 knockdown also notably promoted CRC cell proliferation, metastasis, invasion, and colony formation. To summarize, these experimental results showed that LRRC75A-AS1 might serve as an anti-oncogene for CRC tumorigenesis and advancement, and it may become a novel molecular marker for clinical diagnosis. Impact statement It is reported that colorectal cancer has seriously threatened human health. The incidence of colorectal cancer in China is increasing year by year. At present, the treatment of cancer is gradually developing towards individualized treatment whose core is targeted therapy, and molecular pathology is the basis of targeted therapy. Previous studies have shown that in addition to protein-coding genes that regulate tumor invasion and metastasis, there are also some non-coding genes involved in tumor encroachment and spread. Our study found that long noncoding RNA LRRC75A-AS1 is closely related to CRC and is related to its proliferation and migration. And it may become a novel molecular marker for clinical diagnosis and treatment.

Breast cancer is one of the most frequently diagnosed cancers and the leading cause of cancer death in women. MIER3 (Mesoderm induction early response 1, family member3) is considered as a potential oncogene for breast cancer. However, the role of MIER3 in breast cancer remain largely unknown. The expression of MIER3 was detected and the relationship between its expression and clinicopathological characteristics was also analyzed. The effect of MIER3 on proliferation and migration of breast cancer cells was detected in vitro and in vivo. Western blot, IF, and Co-IP were employed to detect the relationship between MIER3, HDAC1, HDAC2, and Snail. ChIP assay was performed to determine the binding of MIER3/HDAC1/HDAC2/Snail complex to the promoter of E-cadherin. In this study, we found that MIER3 was upregulated in breast cancer tissue and closely associated with poor prognosis of patients. MIER3 could promote the proliferation, migration, and epithelial-mesenchymal transition (EMT) of breast cancer cells. Further studies showed that MIER3 interacted with HDAC1/HDAC2 and Snail to form a repressive complex which could bind to E-cadherin promoter and was related to its deacetylation. Our study concluded that MIER3 was involved in forming a co-repressor complex with HDAC1/HDAC2/Snail to promote EMT by silencing E-cadherin.

While humans can effortlessly transform complex visual scenes into simple words and the other way around by leveraging their high-level understanding of the content, conventional or the more recent learned image compression codecs do not seem to utilize the semantic meanings of visual content to their full potential. Moreover, they focus mostly on rate-distortion and tend to underperform in perception quality especially in low bitrate regime, and often disregard the performance of downstream computer vision algorithms, which is a fast-growing consumer group of compressed images in addition to human viewers. In this paper, we (1) present a generic framework that can enable any image codec to leverage high-level semantics and (2) study the joint optimization of perception quality and distortion. Our idea is that given any codec, we utilize high-level semantics to augment the low-level visual features extracted by it and produce essentially a new, semantic-aware codec. We propose a three-phase training scheme that teaches semantic-aware codecs to leverage the power of semantic to jointly optimize rate-perception-distortion (R-PD) performance. As an additional benefit, semantic-aware codecs also boost the performance of downstream computer vision algorithms. To validate our claim, we perform extensive empirical evaluations and provide both quantitative and qualitative results., Comment: Accepted by IEEE Transactions on Image Processing

Long non-coding RNAs (lncRNAs) have been demonstrated to serve important roles in the development and progression of cancer. Recently HNF1A antisense RNA 1 (HNF1A‑AS1), a lncRNA, has been reported as exhibiting a potential oncogenic role in the development of many types of cancer. However, the expression and the role of HNF1A‑AS1 in colorectal carcinoma (CRC) remains unclear. In the present study, the role of HNF1A‑AS1 in CRC was examined for the first time and its correlation with CRC cell biological behaviors was analyzed. The results demonstrated that HNF1A‑AS1 was distinctly upregulated in CRC tissues and associated with CRC metastasis to the lymph nodes. Reverse transcription‑quantitative polymerase chain reaction revealed that HNF1A‑AS1 was also upregulated in CRC cell lines and localized in the nucleus. In addition, knockdown of HNF1A‑AS1 expression notably inhibited CRC cell proliferation, migration, invasion and colony formation, and suppressed S‑phase entry in vitro. Taken together, these results suggested that HNF1A‑AS1 might serve as a promising prognostic marker for CRC tumorigenesis and progression.

Current studies have shown that POTE ankyrin domain family members have high expressions as tumor antigens in malignant tumors, such as prostate cancer, ovarian cancer, breast cancer and the like. POTEE is a member of the POTE anchor protein family E. However, its role in colorectal carcinoma (CRC) has not been studied. In this study, the function of POTEE in CRC was examined for the first time and its correlation with CRC cell biological behaviors was analyzed. Quantitative reverse transcription-polymerase chain reaction (qRT-PCR), western blotting, and immunohistochemistry revealed that POTEE was remarkably overexpressed in CRC and associated with an aggressive phenotype. We also found that POTEE was localized in the cytoplasm. In addition, downregulation of POTEE expression can notably inhibit the proliferation, migration, and invasion of CRC cell in vitro, and repressed tumor growth and metastasis in vivo. In contrast, overexpression of POTEE could promote the aggressive behaviors of CRC cells. Mechanistically, POTEE promoted CRC migration, invasion and epithelial-mesenchymal transition (EMT) by increasing the activation of Rac1 and Cdc42. To summarize, these results suggested that POTEE might serve as an oncogene for CRC tumorigenesis and progression, and may become a novel molecular marker for clinical diagnosis and treatment.

Following the publication of the article, the authors noted an error associated with the presentation of Fig. 3B. This Figure part showed the cell cycle analysis as determined by flow cytometry of HT29 cells transfected with either a HNF1A-AS1-specific siRNA or a negative control siRNA at 48 h post-transfection. An error was made in the compilation of this Figure, and the same data were inadvertently selected to represent the cell cycle analysis of both the HT29-SI and the HT29-NC cells (i.e., the data relating to the HT29-NC cells were included in the Figure twice). A corrected version of Fig. 3 is shown on the next page. Note that this correction affects neither the interpretation of the data nor the reported conclusions of this work. The authors all agree to this Corrigendum, and regret that this error went unnoticed during the proofs correction stage. They also wish to thank the Editor for allowing them the opportunity to publish this Corrigendum, and regret any inconvenience this error has caused. [the original article was published in Molecular Medicine Reports 16: 4694-4700, 2017; DOI: 10.3892/mmr.2017.7175].

As the study for the modernization of Traditional Chinese Medicine (TCM) is moving continuously forward, a growing bond exists between TCM and modern information processing technology. The determination of syndrome and the study for the compatibility of medicines in TCM are main parts of it. In this paper, we clustered syndromes of lung cancer patients according to the clinical based cases by adopting Clustering Algorithm Based On Sparse Feature Vector algorithm (CABOSFV) algorithm and concluded three TCM classifications for lung cancer. Moreover, by the further study of the compatibility of medicines, numerous matches for critical medicines were proposed, and the results are correspond to clinical data.