Astrocytic TIMP-1 regulates production of Anastellin, an inhibitor of oligodendrocyte differentiation and FTY720 responses.

Astrocyte activation is associated with neuropathology and the production of tissue inhibitor of metalloproteinase-1 (TIMP1). TIMP1 is a pleiotropic extracellular protein that functions both as a protease inhibitor and as a growth factor. Astrocytes that lack expression of Timp1 do not support rat oligodendrocyte progenitor cell (rOPC) differentiation, and adult global Timp1 knockout (Timp1KO) mice do not efficiently remyelinate following a demyelinating injury. Here, we performed an unbiased proteomic analysis and identified a fibronectin-derived peptide called Anastellin (Ana) that was unique to the Timp1KO astrocyte secretome. Ana was found to block rOPC differentiation in vitro and enhanced the inhibitory influence of fibronectin on rOPC differentiation. Ana is known to act upon the sphingosine-1-phosphate receptor 1, and we determined that Ana also blocked the pro-myelinating effect of FTY720 (or fingolimod) on rOPC differentiation in vitro. Administration of FTY720 to wild-type C57BL/6 mice during MOG35-55-experimental autoimmune encephalomyelitis ameliorated clinical disability while FTY720 administered to mice lacking expression of Timp1 (Timp1KO) had no effect. Analysis of Timp1 and fibronectin (FN1) transcripts from primary human astrocytes from healthy and multiple sclerosis (MS) donors revealed lower TIMP1 expression was coincident with elevated FN1 in MS astrocytes. Last, analyses of proteomic databases of MS samples identified Ana peptides to be more abundant in the cerebrospinal fluid (CSF) of human MS patients with high disease activity. A role for Ana in MS as a consequence of a lack of astrocytic TIMP-1 production could influence both the efficacy of fingolimod responses and innate remyelination potential in the MS brain.

Defective sphingosine 1-phosphate receptor 1 (S1P1) phosphorylation exacerbates TH17-mediated autoimmune neuroinflammation.

Sphingosine 1-phosphate (S1P) signaling regulates lymphocyte egress from lymphoid organs into systemic circulation. The sphingosine phosphate receptor 1 (S1P1) agonist FTY-720 (Gilenya) arrests immune trafficking and prevents multiple sclerosis (MS) relapses. However, alternative mechanisms of S1P-S1P1 signaling have been reported. Phosphoproteomic analysis of MS brain lesions revealed S1P1 phosphorylation on S351, a residue crucial for receptor internalization. Mutant mice harboring an S1pr1 gene encoding phosphorylation-deficient receptors (S1P1(S5A)) developed severe experimental autoimmune encephalomyelitis (EAE) due to autoimmunity mediated by interleukin 17 (IL-17)-producing helper T cells (TH17 cells) in the peripheral immune and nervous system. S1P1 directly activated the Jak-STAT3 signal-transduction pathway via IL-6. Impaired S1P1 phosphorylation enhances TH17 polarization and exacerbates autoimmune neuroinflammation. These mechanisms may be pathogenic in MS.

Cross-Referencing Self-Training Network for Sound Event Detection in Audio Mixtures.

Sound event detection is an important facet of audio tagging that aims to identify sounds of interest and define both the sound category and time boundaries for each sound event in a continuous recording. With advances in deep neural networks, there has been tremendous improvement in the performance of sound event detection systems, although at the expense of costly data collection and labeling efforts. In fact, current state-of-the-art methods employ supervised training methods that leverage large amounts of data samples and corresponding labels in order to facilitate identification of sound category and time stamps of events. As an alternative, the current study proposes a semi-supervised method for generating pseudo-labels from unsupervised data using a student-teacher scheme that balances self-training and cross-training. Additionally, this paper explores post-processing which extracts sound intervals from network prediction, for further improvement in sound event detection performance. The proposed approach is evaluated on sound event detection task for the DCASE2020 challenge. The results of these methods on both "validation" and "public evaluation" sets of DESED database show significant improvement compared to the state-of-the art systems in semi-supervised learning.

Extracellular vesicle fibrinogen induces encephalitogenic CD8+ T cells in a mouse model of multiple sclerosis.

Extracellular vesicles (EVs) are emerging as potent mediators of intercellular communication with roles in inflammation and disease. In this study, we examined the role of EVs from blood plasma (pEVs) in an experimental autoimmune encephalomyelitis mouse model of central nervous system demyelination. We determined that pEVs induced a spontaneous relapsing-remitting disease phenotype in MOG35-55-immunized C57BL/6 mice. This modified disease phenotype was found to be driven by CD8+ T cells and required fibrinogen in pEVs. Analysis of pEVs from relapsing-remitting multiple sclerosis patients also identified fibrinogen as a significant portion of pEV cargo. Together, these data suggest that fibrinogen in pEVs contributes to the perpetuation of neuroinflammation and relapses in disease.

Fusion-ConvBERT: Parallel Convolution and BERT Fusion for Speech Emotion Recognition.

Speech emotion recognition predicts the emotional state of a speaker based on the person's speech. It brings an additional element for creating more natural human-computer interactions. Earlier studies on emotional recognition have been primarily based on handcrafted features and manual labels. With the advent of deep learning, there have been some efforts in applying the deep-network-based approach to the problem of emotion recognition. As deep learning automatically extracts salient features correlated to speaker emotion, it brings certain advantages over the handcrafted-feature-based methods. There are, however, some challenges in applying them to the emotion recognition problem, because data required for properly training deep networks are often lacking. Therefore, there is a need for a new deep-learning-based approach which can exploit available information from given speech signals to the maximum extent possible. Our proposed method, called "Fusion-ConvBERT", is a parallel fusion model consisting of bidirectional encoder representations from transformers and convolutional neural networks. Extensive experiments were conducted on the proposed model using the EMO-DB and Interactive Emotional Dyadic Motion Capture Database emotion corpus, and it was shown that the proposed method outperformed state-of-the-art techniques in most of the test configurations.

Integrative proteomics, genomics, and translational immunology approaches reveal mutated forms of Proteolipid Protein 1 (PLP1) and mutant-specific immune response in multiple sclerosis.

In order to gain mechanistic insights into multiple sclerosis (MS) pathogenesis, we utilized a multi-dimensional approach to test the hypothesis that mutations in myelin proteins lead to immune activation and central nervous system autoimmunity in MS. Mass spectrometry-based proteomic analysis of human MS brain lesions revealed seven unique mutations of PLP1; a key myelin protein that is known to be destroyed in MS. Surprisingly, in-depth genomic analysis of two MS patients at the genomic DNA and mRNA confirmed mutated PLP1 in RNA, but not in the genomic DNA. Quantification of wild type and mutant PLP RNA levels by qPCR further validated the presence of mutant PLP RNA in the MS patients. To seek evidence linking mutations in abundant myelin proteins and immune-mediated destruction of myelin, specific immune response against mutant PLP1 in MS patients was examined. Thus, we have designed paired, wild type and mutant peptide microarrays, and examined antibody response to multiple mutated PLP1 in sera from MS patients. Consistent with the idea of different patients exhibiting unique mutation profiles, we found that 13 out of 20 MS patients showed antibody responses against specific but not against all the mutant-PLP1 peptides. Interestingly, we found mutant PLP-directed antibody response against specific mutant peptides in the sera of pre-MS controls. The results from integrative proteomic, genomic, and immune analyses reveal a possible mechanism of mutation-driven pathogenesis in human MS. The study also highlights the need for integrative genomic and proteomic analyses for uncovering pathogenic mechanisms of human diseases.

Discovery of putative breast cancer antigens using an integrative platform of genomics-driven immunoproteomics.

Recent advances in cancer immuno-therapeutics such as checkpoint inhibitors, chimeric antigen-receptor T cells, and tumor infiltrating T cells (TIL) are now significantly impacting cancer patients in a positive manner. Although very promising, reports indicate no more than 25% of cases result in complete remission. One of the limitations of these treatments is the identity of putative cancer antigens in each patient, as it is technically challenging to identify cancer antigens in a rapid fashion. Thus, identification of cancer antigens followed by targeted treatment will increase the efficacy of cancer immunotherapies. To achieve this goal, a combined technologies platform of deep genomic sequencing and personalized immune assessment was devised, termed Genomics Driven Immunoproteomics (GDI). Using this technological platform, we report the discovery of 149 tumor antigens from human breast cancer patients. Significant number of these putative cancer antigens arise from single nucleotide variants (SNVs), as well as insertions and deletions that results into frame-shift mutations. We propose a general model of anti-cancer immunity and suggest that the GDI platform may help identify patient-specific tumor antigens in a timely fashion for precision immunotherapies.

Online multi-object tracking with efficient track drift and fragmentation handling.

This paper addresses the problem of multi-object tracking in complex scenes by a single, static, uncalibrated camera. Tracking-by-detection is a widely used approach for multi-object tracking. Challenges still remain in complex scenes, however, when this approach has to deal with occlusions, unreliable detections (e.g., inaccurate position/size, false positives, or false negatives), and sudden object motion/appearance changes, among other issues. To handle these problems, this paper presents a novel online multi-object tracking method, which can be fully applied to real-time applications. First, an object tracking process based on frame-by-frame association with a novel affinity model and an appearance update that does not rely on online learning is proposed to effectively and rapidly assign detections to tracks. Second, a two-stage drift handling method with novel track confidence is proposed to correct drifting tracks caused by the abrupt motion change of objects under occlusion and prolonged inaccurate detections. In addition, a fragmentation handling method based on a track-to-track association is proposed to solve the problem in which an object trajectory is broken into several tracks due to long-term occlusions. Based on experimental results derived from challenging public data sets, the proposed method delivers an impressive performance compared with other state-of-the-art methods. Furthermore, additional performance analysis demonstrates the effect and usefulness of each component of the proposed method.

Tyrosine phosphorylation of CD13 regulates inflammatory cell-cell adhesion and monocyte trafficking.

CD13 is a large cell surface peptidase expressed on the monocytes and activated endothelial cells that is important for homing to and resolving the damaged tissue at sites of injury. We showed previously that cross-linking of human monocytic CD13 with activating Abs induces strong adhesion to endothelial cells in a tyrosine kinase- and microtubule-dependent manner. In the current study, we examined the molecular mechanisms underlying these observations in vitro and in vivo. We found that cross-linking of CD13 on U937 monocytic cells induced phosphorylation of a number of proteins, including Src, FAK, and ERK, and inhibition of these abrogated CD13-dependent adhesion. We found that CD13 itself was phosphorylated in a Src-dependent manner, which was an unexpected finding because its 7-aa cytoplasmic tail was assumed to be inert. Furthermore, CD13 was constitutively associated with the scaffolding protein IQGAP1, and CD13 cross-linking induced complex formation with the actin-binding protein α-actinin, linking membrane-bound CD13 to the cytoskeleton, further supporting CD13 as an inflammatory adhesion molecule. Mechanistically, mutation of the conserved CD13 cytoplasmic tyrosine to phenylalanine abrogated adhesion; Src, FAK, and ERK phosphorylation; and cytoskeletal alterations upon Ab cross-linking. Finally, CD13 was phosphorylated in isolated murine inflammatory peritoneal exudate cells, and adoptive transfer of monocytic cell lines engineered to express the mutant CD13 were severely impaired in their ability to migrate into the inflamed peritoneum, confirming that CD13 phosphorylation is relevant to inflammatory cell trafficking in vivo. Therefore, this study identifies CD13 as a novel, direct activator of intracellular signaling pathways in pathophysiological conditions.

Large-scale proteomic characterization of melanoma expressed proteins reveals nestin and vimentin as biomarkers that can potentially distinguish melanoma subtypes.

Melanoma is an aggressive type of skin cancer, which accounts for only 4% of skin cancer cases but causes around 75% of skin cancer deaths. Currently, there is a limited set of protein biomarkers that can distinguish melanoma subtypes and provide an accurate prognosis of melanoma. Thus, we have selected and profiled the proteomes of five different melanoma cell lines from different stages of progression in comparison with a normal melanocytes using tandem mass spectrometry. We also profiled the proteome of a solid metastatic melanoma tumor. This resulted in the identification of 4758 unique proteins, among which ∼200-300 differentially expressed proteins from each set were found by quantitative proteomics. Correlating protein expression with aggressiveness of each melanoma cell line and literature mining resulted in the final selection of six proteins: vimentin, nestin, fibronectin, annexin A1, dipeptidyl peptidase IV, and histone H2A1B. Validation of nestin and vimentin using 40 melanoma samples revealed pattern of protein expression can help predict melanoma aggressiveness in different subgroups of melanoma. These results, together with the combined list of 4758 expressed proteins, provide a valuable resource for selecting melanoma biomarkers in the future for the clinical and research community.

Proteomic analysis of active multiple sclerosis lesions reveals therapeutic targets.

Understanding the neuropathology of multiple sclerosis (MS) is essential for improved therapies. Therefore, identification of targets specific to pathological types of MS may have therapeutic benefits. Here we identify, by laser-capture microdissection and proteomics, proteins unique to three major types of MS lesions: acute plaque, chronic active plaque and chronic plaque. Comparative proteomic profiles identified tissue factor and protein C inhibitor within chronic active plaque samples, suggesting dysregulation of molecules associated with coagulation. In vivo administration of hirudin or recombinant activated protein C reduced disease severity in experimental autoimmune encephalomyelitis and suppressed Th1 and Th17 cytokines in astrocytes and immune cells. Administration of mutant forms of recombinant activated protein C showed that both its anticoagulant and its signalling functions were essential for optimal amelioration of experimental autoimmune encephalomyelitis. A proteomic approach illuminated potential therapeutic targets selective for specific pathological stages of MS and implicated participation of the coagulation cascade.

Voice activity detection in noisy environments based on double-combined fourier transform and line fitting.

A new voice activity detector for noisy environments is proposed. In conventional algorithms, the endpoint of speech is found by applying an edge detection filter that finds the abrupt changing point in a feature domain. However, since the frame energy feature is unstable in noisy environments, it is difficult to accurately find the endpoint of speech. Therefore, a novel feature extraction algorithm based on the double-combined Fourier transform and envelope line fitting is proposed. It is combined with an edge detection filter for effective detection of endpoints. Effectiveness of the proposed algorithm is evaluated and compared to other VAD algorithms using two different databases, which are AURORA 2.0 database and SITEC database. Experimental results show that the proposed algorithm performs well under a variety of noisy conditions.

Reinforced AdaBoost learning for object detection with local pattern representations.

A reinforced AdaBoost learning algorithm is proposed for object detection with local pattern representations. In implementing AdaBoost learning, the proposed algorithm employs an exponential criterion as a cost function and Newton's method for its optimization. In particular, we introduce an optimal selection of weak classifiers minimizing the cost function and derive the reinforced predictions based on a judicial confidence estimate to determine the classification results. The weak classifier of the proposed method produces real-valued predictions while that of the conventional AdaBoost method produces integer valued predictions of +1 or -1. Hence, in the conventional learning algorithms, the entire sample weights are updated by the same rate. On the contrary, the proposed learning algorithm allows the sample weights to be updated individually depending on the confidence level of each weak classifier prediction, thereby reducing the number of weak classifier iterations for convergence. Experimental classification performance on human face and license plate images confirm that the proposed method requires smaller number of weak classifiers than the conventional learning algorithm, resulting in higher learning and faster classification rates. An object detector implemented based on the proposed learning algorithm yields better performance in field tests in terms of higher detection rate with lower false positives than that of the conventional learning algorithm.

Astrocytic TIMP-1 regulates production of Anastellin, a novel inhibitor of oligodendrocyte differentiation and FTY720 responses.

Astrocyte activation is associated with neuropathology and the production of tissue inhibitor of metalloproteinase-1 (TIMP1). TIMP1 is a pleiotropic extracellular protein that functions both as a protease inhibitor and as a growth factor. We have previously demonstrated that murine astrocytes that lack expression of Timp1 do not support rat oligodendrocyte progenitor cell (rOPC) differentiation, and adult global Timp1 knockout ( Timp1 KO ) mice do not efficiently remyelinate following a demyelinating injury. To better understand the basis of this, we performed unbiased proteomic analyses and identified a fibronectin-derived peptide called anastellin that is unique to the murine Timp1 KO astrocyte secretome. Anastellin was found to block rOPC differentiation in vitro and enhanced the inhibitory influence of fibronectin on rOPC differentiation. Anastellin is known to act upon the sphingosine-1-phosphate receptor 1 (S1PR1), and we determined that anastellin also blocked the pro-myelinating effect of FTY720 (or fingolimod) on rOPC differentiation in vitro . Further, administration of FTY720 to wild-type C57BL/6 mice during MOG 35-55 -EAE ameliorated clinical disability while FTY720 administered to mice lacking expression of Timp1 in astrocytes ( Timp1 cKO ) had no effect. Analysis of human TIMP1 and fibronectin ( FN1 ) transcripts from healthy and multiple sclerosis (MS) patient brain samples revealed an inverse relationship where lower TIMP1 expression was coincident with elevated FN1 in MS astrocytes. Lastly, we analyzed proteomic databases of MS samples and identified anastellin peptides to be more abundant in the cerebrospinal fluid (CSF) of human MS patients with high versus low disease activity. The prospective role for anastellin generation in association with myelin lesions as a consequence of a lack of astrocytic TIMP-1 production could influence both the efficacy of fingolimod responses and the innate remyelination potential of the the MS brain. Significance Statement: Astrocytic production of TIMP-1 prevents the protein catabolism of fibronectin. In the absence of TIMP-1, fibronectin is further digested leading to a higher abundance of anastellin peptides that can bind to sphingosine-1-phosphate receptor 1. The binding of anastellin with the sphingosine-1-phosphate receptor 1 impairs the differentiation of oligodendrocytes progenitor cells into myelinating oligodendrocytes in vitro , and negates the astrocyte-mediated therapeutic effects of FTY720 in the EAE model of chronic CNS inflammation. These data indicate that TIMP-1 production by astrocytes is important in coordinating astrocytic functions during inflammation. In the absence of astrocyte produced TIMP-1, elevated expression of anastellin may represent a prospective biomarker for FTY720 therapeutic responsiveness.

LC-MS/MS Identification of a Bromelain Peptide Biomarker from Ananas comosus Merr.

Bromelain (Br) is a cysteine peptidase (GenBank AEH26024.1) from pineapple, with over 40 years of clinical use. The constituents mediating its anti-inflammatory activity are not thoroughly characterized and no peptide biomarker exists. Our objective is to characterize Br raw material and identify peptides in the plasma of Br treated mice. After SDS-PAGE in-gel digestion, Br (VN#3507; Middletown, CT, USA) peptides were analyzed via LC/MS/MS using 95% protein probability, 95% peptide probability, and a minimum peptide number = 5. Br spiked mouse plasma (1 ug/ul) and plasma from i.p. treated mice (12 mg/kg) were assessed using SRM. In Br raw material, we identified seven proteins: four proteases, one jacalin-like lectin, and two protease inhibitors. In Br spiked mouse plasma, six proteins (ananain, bromelain inhibitor, cysteine proteinase AN11, FB1035 precursor, FBSB precursor, and jacalin-like lectin) were identified. Using LC/MS/MS, we identified the unique peptide, DYGAVNEVK, derived from FB1035, in the plasma of i.p. Br treated mice. The spectral count of this peptide peaked at 6 hrs and was undetectable by 24 hrs. In this study, a novel Br peptide was identified in the plasma of treated mice for the first time. This Br peptide could serve as a biomarker to standardize the therapeutic dose and maximize clinical utility.

Human myelin proteome and comparative analysis with mouse myelin.

Myelin, formed by oligodendrocytes (OLs) in the CNS, is critical for axonal functions, and its damage leads to debilitating neurological disorders such as multiple sclerosis. Understanding the molecular mechanisms of myelination and the pathogenesis of human myelin disease has been limited partly by the relative lack of identification and functional characterization of the repertoire of human myelin proteins. Here, we present a large-scale analysis of the myelin proteome, using the shotgun approach of 1-dimensional PAGE and liquid chromatography/tandem MS. Three hundred eight proteins were commonly identified from human and mouse myelin fractions. Comparative microarray analysis of human white and gray matter showed that transcripts of several of these were elevated in OL-rich white matter compared with gray matter, providing confidence in their detection in myelin. Comparison with other databases showed that 111 of the identified proteins/transcripts also were expressed in OLs, rather than in astrocytes or neurons. Comparison with 4 previous myelin proteomes further confirmed more than 50% of the identified proteins and revealed the presence of 163 additional proteins. A select group of identified proteins also were verified by immunoblotting. We classified the identified proteins into biological subgroups and discussed their relevance in myelin biogenesis and maintenance. Taken together, the study provides insights into the complexity of this metabolically active membrane and creates a valuable resource for future in-depth study of specific proteins in myelin with relevance to human demyelinating diseases.

Subcellular tissue proteomics of hepatocellular carcinoma for molecular signature discovery.

Hepatocellular carcinoma (HCC) is one of the leading causes of mortality from solid organ malignancy worldwide. Because of the complexity of proteins within liver cells and tissues, the discovery of therapeutic targets of HCC has been difficult. To investigate strategies for decreasing the complexity of tissue samples for detecting meaningful protein mediators of HCC, we employed subcellular fractionation combined with 1D-gel electrophoresis and liquid chromatography-tandem mass spectrometry analysis. Moreover, we utilized a statistical method, namely, the Power Law Global Error Model (PLGEM), to distinguish differentially expressed proteins in a duplicate proteomic data set. Mass spectrometric analysis identified 3045 proteins in nontumor and HCC from cytosolic, membrane, nuclear, and cytoskeletal fractions. The final lists of highly differentiated proteins from the targeted fractions were searched for potentially translocated proteins in HCC from soluble compartments to the nuclear or cytoskeletal compartments. This analysis refined our targets of interest to include 21 potential targets of HCC from these fractions. Furthermore, we validated the potential molecular targets of HCC, MATR3, LETM1, ILF2, and IQGAP2 by Western blotting, immunohistochemisty, and immunofluorescent microscopy. Here we demonstrate an efficient strategy of subcellular tissue proteomics toward molecular target discovery of one of the most complicated human disease, HCC.

Sphingosine interaction with acidic leucine-rich nuclear phosphoprotein-32A (ANP32A) regulates PP2A activity and cyclooxygenase (COX)-2 expression in human endothelial cells.

Sphingolipid metabolites regulate cell fate by acting on specific cellular targets. Although the influence of sphingolipids in cellular signaling has been well recognized, the exact molecular targets and how these targets influence cellular signaling mechanisms remain poorly understood. Toward this goal, we used affinity chromatography coupled with proteomics technology and identified acidic leucine-rich nuclear phosphoprotein-32A (ANP32A), an inhibitor of protein phosphatase 2A (PP2A) as a direct target of sphingosine, N,N'-dimethyl sphingosine (DMS) and phytosphingosine but not dihydrosphingosine or sphingosine 1-phosphate. Treatment of human umbilical vein endothelial cells (HUVEC) with DMS, which is not phosphorylated by sphingosine kinases, led to the activation of PP2A activity. Suppression of ANP32A with siRNA enhanced basal and DMS-activated PP2A activity suggesting that the sphingoid base binds to and relieves the inhibitory action of ANP32A on the PP2A complex. Indeed, DMS relieved the ANP32A-mediated inhibition of PP2A enzyme complex in vitro. Interestingly, DMS treatment induced the p38 stress-activated protein kinase (SAPK) and expression of cyclooxygenase (COX)-2 transcript and protein. Knockdown of ANP32A expression further induced p38 SAPK and COX-2. These data identify ANP32A as a novel molecular target of sphingoid bases that regulates cellular signaling events and inflammatory gene expression.

Role of spectral counting in quantitative proteomics.

Spectral count, defined as the total number of spectra identified for a protein, has gained acceptance as a practical, label-free, semiquantitative measure of protein abundance in proteomic studies. In this review, we discuss issues affecting the performance of spectral counting relative to other label-free methods, as well as its limitations. Possible consequences of modifications, which are commonly applied to raw spectral counts to improve abundance estimations, are considered. The use of spectral counting for different types of quantitation studies is explored and critiqued. Different statistical methods and underlying frameworks that have been applied to spectral count analysis are described and compared, and problem areas that undermine confident statistical analysis are considered. Finally, the issue of accurate estimation of false-discovery rates is addressed and identified as a major current challenge in quantitative proteomics.

Fusion of Heterogeneous Adversarial Networks for Single Image Dehazing.

In this paper, we propose a novel image dehazing method. Typical deep learning models for dehazing are trained on paired synthetic indoor dataset. Therefore, these models may be effective for indoor image dehazing but less so for outdoor images. We propose a heterogeneous Generative Adversarial Networks (GAN) based method composed of a cycle-consistent Generative Adversarial Networks (CycleGAN) for producing haze-clear images and a conditional Generative Adversarial Networks (cGAN) for preserving textural details. We introduce a novel loss function in the training of the fused network to minimize GAN generated artifacts, to recover fine details, and to preserve color components. These networks are fused via a convolutional neural network (CNN) to generate dehazed image. Extensive experiments demonstrate that the proposed method significantly outperforms the state-of-the-art methods on both synthetic and real-world hazy images.