Bridging the Gap of AutoGraph Between Academia and Industry: Analyzing AutoGraph Challenge at KDD Cup 2020.

Graph structured data is ubiquitous in daily life and scientific areas and has attracted increasing attention. Graph Neural Networks (GNNs) have been proved to be effective in modeling graph structured data and many variants of GNN architectures have been proposed. However, much human effort is often needed to tune the architecture depending on different datasets. Researchers naturally adopt Automated Machine Learning on Graph Learning, aiming to reduce human effort and achieve generally top-performing GNNs, but their methods focus more on the architecture search. To understand GNN practitioners' automated solutions, we organized AutoGraph Challenge at KDD Cup 2020, emphasizing automated graph neural networks for node classification. We received top solutions, especially from industrial technology companies like Meituan, Alibaba, and Twitter, which are already open sourced on GitHub. After detailed comparisons with solutions from academia, we quantify the gaps between academia and industry on modeling scope, effectiveness, and efficiency, and show that (1) academic AutoML for Graph solutions focus on GNN architecture search while industrial solutions, especially the winning ones in the KDD Cup, tend to obtain an overall solution (2) with only neural architecture search, academic solutions achieve on average 97.3% accuracy of industrial solutions (3) academic solutions are cheap to obtain with several GPU hours while industrial solutions take a few months' labors. Academic solutions also contain much fewer parameters.

Codabench: Flexible, easy-to-use, and reproducible meta-benchmark platform.

Obtaining a standardized benchmark of computational methods is a major issue in data-science communities. Dedicated frameworks enabling fair benchmarking in a unified environment are yet to be developed. Here, we introduce Codabench, a meta-benchmark platform that is open sourced and community driven for benchmarking algorithms or software agents versus datasets or tasks. A public instance of Codabench is open to everyone free of charge and allows benchmark organizers to fairly compare submissions under the same setting (software, hardware, data, algorithms), with custom protocols and data formats. Codabench has unique features facilitating easy organization of flexible and reproducible benchmarks, such as the possibility of reusing templates of benchmarks and supplying compute resources on demand. Codabench has been used internally and externally on various applications, receiving more than 130 users and 2,500 submissions. As illustrative use cases, we introduce four diverse benchmarks covering graph machine learning, cancer heterogeneity, clinical diagnosis, and reinforcement learning.

Tributyltin induces disruption of microfilament in HL7702 cells via MAPK-mediated hyperphosphorylation of VASP.

Tributyltin (TBT) has been widely used for various industrial purposes, and it has toxic effects on multiple organs and tissues. Previous studies have found that TBT could induce cytoskeletal disruption, especially of the actin filaments. However, the underlying mechanisms remain unclear. The aim of the present study was to determine whether TBT could induce microfilament disruption using HL7702 cells and then to assess for the total levels of various microfilament-associated proteins; finally, the involvement of the MAPK pathway was investigated. The results showed that after TBT treatment, F-actin began to depolymerize and lost its characteristic filamentous structure. The protein levels of Ezrin and Cofilin remained unchanged, the actin-related protein (ARP) 2/3 levels decreased slightly, and the vasodilator-stimulated phosphoprotein (VASP) decreased dramatically. However, the phosphorylation levels of VASP increased 2.5-fold, and the ratio of phosphorylated-VASP/unphosphorylated-VASP increased 31-fold. The mitogen-activated protein kinases (MAPKs) ERK and JNK were discovered to be activated. Inhibition of ERK and JNK not only largely diminished the TBT-induced hyperphosphorylation of VASP but also recovered the cellular morphology and rescued the cells from death. In summary, this study demonstrates that TBT-induced disruption of actin filaments is caused by the hyperphosphorylation of VASP through MAPK pathways. © 2015 Wiley Periodicals, Inc. Environ Toxicol 31: 1530-1538, 2016.

Mechanisms of microcystin-LR-induced cytoskeletal disruption in animal cells.

Microcystin-LR (MC-LR), a potent hepatotoxin produced by certain bloom-forming cyanobacteria, covalently binds to serine/threonine protein phosphatases and acts as an efficient inhibitor of this group of enzymes. MC-LR induces oxidative stress and the unfolded protein response in multiple cell types, leading to apoptosis through the mitochondrial and endoplasmic reticulum pathways. Histologic lesions of acute MC-LR toxicosis exhibit membrane blebbing, cell rounding and dissociation, indicating that this toxin may exert hepatotoxic effects by causing cytoskeletal disruption. Both in vivo and in vitro studies have revealed that exposure of human, mouse, or rat hepatocytes to MC-LR induces the rearrangement or collapse of the three components of the cytoskeleton. In addition, multiple cytoskeletal and cytoskeleton-associated proteins have been found to be affected by MC-LR. This review summarizes the increasing information in the literature pertaining to the molecular mechanisms of MC-LR-induced cytoskeletal disruption and may increase our understanding of its toxicity.

Hyperphosphorylation of microfilament-associated proteins is involved in microcystin-LR-induced toxicity in HL7702 cells.

Microcystin-LR (MC-LR) has been regarded as a hepatotoxin, which can cause cytoskeletal reorganization, especially of the actin filaments. However, the underlying mechanisms remain unclear. In this study, whether MC-LR could induce microfilaments disruption was verified in the normal human liver cell line HL7702; and then the transcription, translation, and phosphorylation levels of major microfilament-associated proteins were measured; finally, the underlying mechanisms was investigated. After treatment with MC-LR, the actin filaments lost their characteristic filamentous organization in the cells, demonstrating increased actin depolymerization. The mRNA and protein levels of ezrin, vasodilator-stimulated phosphoprotein (VASP), actin-related protein2/3, and cofilin remained unchanged. However, the phosphorylation levels of ezrin and VASP were increased, when treated with 10 µM MC-LR. Moreover, P38 and ERK1/2 were involved in MC-LR-induced hyperphosphorylation of microfilament-associated proteins. In summary, this study demonstrates that MC-LR can cause disruption of actin filaments in HL7702 cells due to MC-LR-induced mitogen-activated protein kinase pathway activation and hyperphosphorylation of different types of microfilament-associated proteins.

Increased expression of TBX2 is a novel independent prognostic biomarker of a worse outcome in colorectal cancer patients after curative surgery and a potential therapeutic target.

T-box2 (TBX2) plays a critical role in embryonic development. Recently, deregulated expression of TBX2 has been implicated in several malignancies. However, the expression and the role of TBX2 in colorectal cancer (CRC) remain unclear. In this study, we found that TBX2 was obviously up-regulated in CRC in comparison with the corresponding normal mucosa at transcriptional and protein level. Up-expression of TBX2 was significantly associated with depth of tumor invasion (P = 0.006), distant metastasis (P = 0.038), advanced AJCC stage (P = 0.008), and relapse (P = 0.003). TBX2 was a significantly prognostic factor for decreased survival and increased disease recurrence independent of tumor stage(II, III stage) and functioned as a biomarker to identify prognosis of patients with CRC (OS: HR 2.154; 95% CI 1.019-4.551; P = 0.044, DFS: HR 2.253; 95% CI 1.109-4.575; P = 0.025). Furthermore, TBX2 could serve as a potential target of cancer drug therapy.

[Advances in genome-wide association study of tuberculosis].

Tuberculosis, caused by Mycobacterium tuberculosis (MTB), is one of the oldest and most influential diseases in the history due to its devastating effect on health and high mortality rate worldwide. Tuberculosis causes more human deaths than any other single infectious disease and the incidence of the tuberculosis is increasing dramatically in recent years. Genome-wide association study (GWAS) has been used to delineate the genetic basis of tuberculosis, and several susceptibility genes and loci were found, which provids important clues to the early intervention and treatment of tuberculosis. However, due to difference in the population structure and host-pathogen interactions, GWAS on tuberculosis faces great challenges. In this review, we introduced the achievements of GWAS on tuberculosis, and illustrated challenges and strategies in the future study.

Identification and validation that up-expression of HOXA13 is a novel independent prognostic marker of a worse outcome in gastric cancer based on immunohistochemistry.

Homeobox (HOX) gene family is known to be classic examples of the intimate relationship between embryogenesis and tumorigenesis. However, less is known about the involvement of HOX gene family with gastric cancerogenesis. Here, we screened the expression of HOX gene family in gastric cancers and explored the relationships between them by cDNA microarray. We found several differentially expressed HOX genes in gastric cancers, especially HOXA10 (11/12) and HOXA13 (11/12) with significantly higher expression in the cancerous tissues. Furthermore, we validated HOXA13 as a novel prognostic marker in gastric cancer based on immunohistochemistry and statistical analysis. HOXA13 expression was significantly up-regulated in cancerous tissues compared with the corresponding non-cancerous mucosa (P < 0.001). Up-expression of HOXA13 was significantly correlated with T stage (P = 0.002), M stage (P = 0.024), advanced UICC stage (P < 0.001), histological differentiation (P = 0.005), and relapse (P = 0.001). Patients with positive HOXA13 expression had a obviously lower overall survival (OS) and disease-free survival (DFS) rate than patients with negative HOXA13 expression (HR 3.331, 95 % CI 1.722-6.442, P < 0.001; HR 3.289, 95 % CI 1.703-6.351, P < 0.001, respectively). Univariate and multivariate Cox analysis confirmed that HOXA13 could serve as a significant independent prognostic factor for DFS and OS. Therefore, our results indicated that several HOX genes might be closely involved in the process of the gastric tumorigenesis. Furthermore, up-expression of HOXA13 might be associated with highly aggressive phenotype of gastric cancer. HOXA13 was a significant independent prognostic factor and could serve as a putative biomarker for diagnosis and prognosis of gastric cancer.

Hyperphosphorylation of intermediate filament proteins is involved in microcystin-LR-induced toxicity in HL7702 cells.

Microcystin-LR (MC-LR) is commonly characterized as a hepatotoxin, which can cause disruption of keratin filaments. Keratins, however, account for only two types of intermediate filaments (IFs), and the potential involvement of other IF proteins in MC-LR-induced toxicity and the underlying mechanisms are still unclear. In this study, the human normal liver cell line HL7702 was used to investigate whether MC-LR can change the transcription, translation, and phosphorylation levels of major IF proteins and to elucidate the underlying mechanisms. The results showed that MC-LR triggered an accumulation of IFs around the nucleus and led to the formation of dense bundles. When the cells were treated with 10µM MC-LR, cell proliferation significantly decreased with an increase in apoptosis and cell cycle arrest. Moreover, the mRNA and protein levels of keratin 18, vimentin and lamin A/C were not changed; however, the phosphorylation of K8/18 and vimentin was significantly increased. Furthermore, we found MC-LR exposure caused phosphoactivation of P38, JNK and ERK1/2 in a concentration-dependent manner, and P38 and ERK1/2 were involved in MC-LR-induced hyperphosphorylation of IF proteins. Taken together, the results of this study suggest that MC-LR exerts its potential hepatotoxicity through MAPK pathway activation, which cause hyperphosphorylation of IF proteins and result in cytoskeletal architecture remodeling and cell survival/death regulation. Since IFs serve as signaling platforms and dozens of IF proteins are involved in different signaling pathways, future studies focus on different IFs may provide helpful insights into the mechanisms of MC-LR toxicity.