Biology subjects

Zhu, W.

Publications and source records attributed to Zhu, W..

187 records · Page 11Linked to original sources

bioRxiv2017

DeepLung: 3D Deep Convolutional Nets for Automated Pulmonary Nodule Detection and Classification

In this work, we present a fully automated lung CT cancer diagnosis system, DeepLung. DeepLung contains two parts, nodule detection and classification. Considering the 3D nature of lung CT data, two 3D networks are designed for the nodule detection and classification respectively. Specifically, a 3D Faster R-CNN is designed for nodule detection with a U-net-like encoder-decoder structure to effectively learn nodule features. For nodule classification, gradient boosting machine (GBM) with 3D dual path network (DPN) features is proposed. The nodule classification subnetwork is validated on a public dataset from LIDC-IDRI, on which it achieves better performance than state-of-the-art approaches, and surpasses the average performance of four experienced doctors. For the DeepLung system, candidate nodules are detected first by the nodule detection subnetwork, and nodule diagnosis is conducted by the classification subnetwork. Extensive experimental results demonstrate the DeepLung is comparable to the experienced doctors both for the nodule-level and patient-level diagnosis on the LIDC-IDRI dataset.

bioinformatics↗

bioRxiv2017

A rainfall-manipulation experiment with 517 Arabidopsis thaliana accessions

The gold standard for studying natural selection and adaptation in the wild is to quantify lifetime fitness of individuals from natural populations that have been grown together in a common garden, or that have been reciprocally transplanted. By combining fitness values with species traits and genome sequences, one can infer selection coefficients at the genetic level. Here we present a rainfall-manipulation experiment with 517 whole-genome sequenced natural accessions of the plant Arabidopsis thaliana spanning the global distribution of the species. The experiments were conducted in two field stations in contrasting climates, in the Mediterranean and in Central Europe, where we built rainout shelters and simulated high and low rainfall. Using custom image analysis we quantified fitness- and phenology-related traits for 23,154 pots, which contained about 14,500 plants growing independently, and over 310,000 plants growing in small populations (max. 30 plants). This large field experiment dataset, which associates fitness and ecologically-relevant traits with genomes, will provide an important resource to test eco-evolutionary genetic theories and to understand the potential evolutionary impacts of future climates on an important plant model species.

evolutionary biology↗

bioRxiv2017

Aging-associated dysbiosis increases susceptibility to enteric viral infection in Drosophila

Age is associated with increased susceptibility to enteric infections, but the molecular mechanisms are unclear. We find that aged Drosophila are more susceptible to enteric viral infections and that this increase in susceptibility is due to the aged microbiota, since depletion of the microbiota or reconstitution with a young microbiome suppressed infection. Metagenomic analysis of the aged microbiome revealed dysbiosis with an increased abundance in reactive oxygen species (ROS) producing pathways. This aged microbiota drives intestinal ROS production and we could restore immune function in old flies by reducing ROS genetically or pharmacologically. Moreover, we found that reconstitution of old flies with a cocktail of commensals, including L. fructivorans and heat-killed A. pomorum, could fully restore immunity. Altogether, these findings provide a mechanistic link between age-dependent dysbiosis and antiviral immunity and show that we can restore innate protection in aged animals, suggesting that this is a treatable and reversible state.

immunology↗

bioRxiv2017

Faster carbon accumulation in global forest soils

Comparing soil organic carbon (SOC) stocks across space and time is a fundamental issue in global ecology. However, the conventional approach fails to determine SOC stock in an equivalent volume of mineral-soil, and therefore, SOC stock changes can be under- or overestimates if soils swell or shrink during forest development or degradation. Here, we propose to estimate SOC stock as the product of mineral-soil mass in an equivalent mineral-soil volume and SOC concentration expressed as g C Kg-1 mineral-soil. This method enables researchers to compare SOC stocks across space and time. Our results show an unaccounted SOC accumulation of 2.4 - 10.1 g C m-2 year-1 in the 1m surface mineral-soils in global forests. This unaccounted SOC amounts to an additional C sink of 0.12 - 0.25 Pg C year-1, which equals 30 - 62% of the previously estimated annual SOC accumulation in global forests. This finding suggests that forest soils are stronger C sinks than previously recognized.

ecology↗

bioRxiv2017

Daam2 Driven Degradation Of VHL Promotes Gliomagenesis

Von Hippel-Landau (VHL) protein is a potent tumor suppressor regulating numerous pathways that drive cancer, but mutations in VHL are restricted to limited subsets of malignancies. Here we identified a novel mechanism for VHL suppression in tumors that do not have inactivating mutations. Using developmental processes to uncover new pathways contributing to tumorigenesis, we found that Daam2 promotes glioma formation. Protein expression screening identified an inverse correlation between Daam2 and VHL expression across a host of cancers, including glioma. These in silico insights guided corroborating functional studies, which revealed that Daam2 promotes tumorigenesis by suppressing VHL expression. Furthermore, biochemical analyses demonstrate that Daam2 associates with VHL and facilitates its ubiquitination and degradation. Together, these studies are the first to define an upstream mechanism regulating VHL suppression in cancer and describe the role of Daam2 in tumorigenesis.\n\nStatement of SignificanceWe found that the glial developmental factor Daam2 promotes glioma tumorigenesis by suppressing VHL expression. Our studies show, for the first time, a regulatory mechanism that operates upstream of VHL in cancer and provides an explanation for how VHL expression is extinguished in tumors that do not have inactivating mutations.

cancer biology↗

bioRxiv2016

Adversarial Deep Structural Networks for Mammographic Mass Segmentation

Mass segmentation is an important task in mammogram analysis, providing effective morphological features and regions of interest (ROI) for mass detection and classification. Inspired by the success of using deep convolutional features for natural image analysis and conditional random fields (CRF) for structural learning, we propose an end-to-end network for mammographic mass segmentation. The network employs a fully convolutional network (FCN) to model potential function, followed by a CRF to perform structural learning. Because the mass distribution varies greatly with pixel position, the FCN is combined with position priori for the task. Due to the small size of mammogram datasets, we use adversarial training to control over-fitting. Four models with different convolutional kernels are further fused to improve the segmentation results. Experimental results on two public datasets, INbreast and DDSM-BCRP, show that our end-to-end network combined with adversarial training achieves the-state-of-the-art results.

bioinformatics↗

bioRxiv2016

Deep Multi-instance Networks with Sparse Label Assignment for Whole Mammogram Classification

Mammogram classification is directly related to computer-aided diagnosis of breast cancer. Traditional methods requires great effort to annotate the training data by costly manual labeling and specialized computational models to detect these annotations during test. Inspired by the success of using deep convolutional features for natural image analysis and multi-instance learning for labeling a set of instances/patches, we propose end-to-end trained deep multiinstance networks for mass classification based on whole mammogram without the aforementioned costly need to annotate the training data. We explore three different schemes to construct deep multi-instance networks for whole mammogram classification. Experimental results on the INbreast dataset demonstrate the robustness of proposed deep networks compared to previous work using segmentation and detection annotations in the training.

bioinformatics↗