Boosting Adversarial Training with Learnable Distribution

In recent years,various adversarial defense methods have been proposed to improve the robustness of deep neural networks.Adversarial training is one of the most potent methods to defend against adversarial attacks.However,the difference in the feature space between natural and adversarial examples hinders the accuracy and robustness of the model in adversarial training.This paper proposes a learnable distribution adversarial training method,aiming to construct the same distribution for training data utilizing the Gaussian mixture model.The distribution centroid is built to classify samples and constrain the distribution of the sample features.The natural and adversarial examples are pushed to the same distribution centroid to improve the accuracy and robustness of the model.The proposed method generates adversarial examples to close the distribution gap between the natural and adversarial examples through an attack algorithm explicitly designed for adversarial training.This algorithm gradually increases the accuracy and robustness of the model by scaling perturbation.Finally,the proposed method outputs the predicted labels and the distance between the sample and the distribution centroid.The distribution characteristics of the samples can be utilized to detect adversarial cases that can potentially evade the model defense.The effectiveness of the proposed method is demonstrated through comprehensive experiments.

Natural forest ALS-TLS point cloud data registration without control points

Airborne laser scanning(ALS)and terrestrial laser scanning(TLS)has attracted attention due to their forest parameter investigation and research applications.ALS is limited to obtaining fi ne structure information below the forest canopy due to the occlusion of trees in natural forests.In contrast,TLS is unable to gather fi ne structure information about the upper canopy.To address the problem of incomplete acquisition of natural forest point cloud data by ALS and TLS on a single platform,this study proposes data registration without control points.The ALS and TLS original data were cropped according to sample plot size,and the ALS point cloud data was converted into relative coordinates with the center of the cropped data as the origin.The same feature point pairs of the ALS and TLS point cloud data were then selected to register the point cloud data.The initial registered point cloud data was fi nely and optimally registered via the iterative closest point(ICP)algorithm.The results show that the proposed method achieved highprecision registration of ALS and TLS point cloud data from two natural forest plots of Pinus yunnanensis Franch.and Picea asperata Mast.which included diff erent species and environments.An average registration accuracy of 0.06 m and 0.09 m were obtained for P.yunnanensis and P.asperata,respectively.

多发肺部磨玻璃样结节EGFR突变状态的研究

目的:探讨同期多发肺部磨玻璃样结节(synchronous multiple ground-glass nodules,SMGGNs)不同病灶间表皮生长因子受体(epidermal growth factor receptor,EGFR)的突变情况并分析其与临床特征的关系。方法:回顾性分析2017年1月至2018年12月河北医科大学第四医院行手术治疗的35例SMGGNs患者的临床病例资料。应用二代测序(next generation sequencing,NGS)技术对手术标本进行检测,检测EGFR基因第18、19、20及21外显子的突变状态。分析病变的EGFR突变状态与患者性别、年龄、病变位置、结节影像学表现和病理学类型的关系。结果:全部患者的EGFR突变率为65.7%(23/35)。女性及不吸烟患者具有更高的EGFR突变率(P=0.015,P<0.001)。浸润性腺癌(ivasive adenocarcinoma,IAC)的EGFR突变率高于非典型腺瘤样增生(atypical ade-nomatoid hyperplasia,AAH)、原位腺癌(adenocarcinoma in situ,AIS)以及微浸润腺癌(minimally ivasive adenocarcinoma,MIA)(P<0.001)。19和21外显子突变为EGFR基因最常见的突变亚型。不同病理亚型的EGFR基因突变状态无显著性差异(P=0.707)。27例具有多个可检测EGFR突变状态结节的患者中,EGFR突变状态的不一致率为85.2%(23/27)。结论:SMGGNs患者不同病灶的EGFR基因突变谱不同,提示各枚结节的发生发展为独立事件。而EGFR基因的突变与磨玻璃样结节(ground glass nodule,GGN)发生发展过程,尤其是肿瘤的侵袭过程密切相关。

Preparation and Characterization of Three Monoclonal Antibodies against HIV-1 p24 Capsid Protein

HIV-1 p24 detection provides a means to aid the early diagnosis of HIV-1 infection, track the progression of disease and assess the efficacy of antiretroviral therapy. In the present study, three monoclonal antibodies （mAbs） p3JB9, p5F1 and p6F4 against HIV-1 p24 were generated. All mAbs could detect p24 of HIV-1ⅢB, HIV-1Ada-M, HIV-174v mAbs p5F1 and p6F4 could detect HIV-1KM018, while p3JB9 could not. Three mAbs did not react with HIV-2ROD, HIV-2CBL-20 and SIVagmTYO-1. The recognized epitope of p5F1 was located on the Gag amino acid region DCKTILKALGPAATLEEMMTAC. The p5F1 was used to establish a modified sandwich ELISA with rabbit anti-p24 serum and showed good specificity and high sensitivity, which has been used to measure HIV-1 p24 antigen levels in research. Cellular ＆ Molecular Immunology.

Immune Responses to Six Synthetic Peptides of Capsid Protein with Sera from HIV-1 Infected Individuals

Many B cell epitopes within p24 of human immunodeficiency virus type 1 （HIV-1） were identified, while most of them were determined by using murine monoclonal antibodies reacting with overlapping peptides of p24. Therefore these epitopes may not represent the actual epitopes recognized by the HIV-1 infected individuals. In the present study, immune responses of 67 HIV-1 positive sera from Yunnan Province, China to five peptides on p24 of HIV-1 and one of HIV-2 were analyzed. All of 67 sera did not recognize peptide GA-12 on HIV-1 and peptide AG-23 on HIV-2, which indicated that GA-12 was not human B cell epitope and AG-23 did not cross-react with HIV-1 positive serum. Except 13 sera （19.4%）, all remaining sera did not recognize peptides NI-15, DR-16, DC-22 and PS-18, which indicated that these four peptides represented B cell linear epitopes of HIV-1 p24 in some HIV-1 infected individuals but not the immuno-dominant epitopes in most individuals.

Dynamic Balance of pSTAT1 and pSTAT3 in C57BL/6 Mice Infected with Lethal or Nonlethal Plasmodium yoelii

Signal transducer and activator of transcription （STAT） proteins play an important role in cytokine signaling pathways and regulation of immune responses. The balance of the phosphorylated （activated） STAT1 （pSTAT1） and STAT3 （pSTAT3） has been documented in cancer immunology. In this study, we investigated the dynamic balance of pSTAT1 and pSTAT3 in C57BL/6 mice infected with either a nonlethal （Py17XNL） or lethal （Py17XL） strain of Plasmodium yoelii. Both Py17XNL and Py17XL infections induced a maximum activation of STAT1 and STAT3 on the first day after parasite inoculation. Additionally, the Py17XNL infection induced a pSTAT1- dominant response in mice during the early stage of infection, with the resolution of parasitemia. In contrast, Py17XL infection induced a pSTAT3-dominant response during the early phase of infection, with the death of the animals. Our results indicated that maximum activation of STAT1 and STAT3 occurred much earlier than the peak levels of cytokines induced by Plasmodium yoelii infection based on previous reports and that infection with Py17XNL and Py17XL induced different dynamic patterns of pSTAT1 and pSTAT3 balance.