基于图像识别和卷积神经网络的大豆优良籽粒筛选研究

为实现通过籽粒图像识别方法对大豆籽粒的品质进行快速、准确检测,以大豆正常品质籽粒及非正常品质籽粒的分类为例,提出一种基于卷积神经网络的大豆优良籽粒图像筛选分类识别方法。建立大豆籽粒品质数据集,设计卷积神经网络,提取大豆籽粒图像特征。为提高分类准确率和实时性,从设计选择卷积神经网络结构、减小过拟合、加快训练收敛速度、增强网络的鲁棒性等方面对卷积神经网络进行优化,最终选择含有4个卷积层、4个池化层、2个全连接层的6层卷积神经网络,采用L2正则化和小批量训练学习方法对网络进行优化训练测试。将结果与传统机器学习分类方法进行比较,试验结果表明:优化的卷积神经网络对大豆籽粒品质分类的准确率达到98.8%,平均检测一幅大豆单籽粒图像的时间为2.96 ms,可为大豆籽粒品质划分提供重要参考。

作物回交导入系的构建与应用

回交导入系是利用回交及标记辅助选择的手段构建而成的遗传与育种材料。经过多代回交,后代材料在轮回亲本的遗传背景下只包含一个或少量供体亲本染色体片段,因此,可作为QTL分析的重要材料。同时,多代回交有利于打破优异基因与不良基因的连锁,优异基因导入到整体表现优良的轮回亲本材料中,进而实现对育种材料的改良。鉴于其一致的遗传背景,导入系在QTL精细定位和基因克隆、QTL间互作研究、遗传验证及作物聚合育种和分子设计育种中均发挥重要的作用。本文对回交导入系的构建及其在作物遗传育种中的应用进行综述。

基于语义分割的大豆主茎精准表型自动获取及其在导入系材料选择中的应用

大豆导入系群体的构建和材料的选择是大豆育种的重要手段之一,而针对某一具体表型的材料选择不仅能够提高育种效率,而且还能够加快定位目标性状基因的进程。本研究针对由绥农14(轮回亲本)和野生野生大豆ZYD00006(供体亲本)为双亲构建的导入系(190个个体),采用机器视觉中的语义分割技术,对主茎相关表型进行了自动提取,实验结果表明:此方法完全可以取代人工测量且具有良好的可移植性,同样可用于其他作物的表型获取。在此基础上,本研究针对大豆主茎相关表型,对参试材料进行系统聚类,结果明确给出各参试材料(近等材料)基于主茎表型的聚类情况。这一结果将成为针对主茎表型进行材料选择的重要依据,同时依据这一结果可以大大加快主茎相关表型基因的定位。

2015—2018年住院患者治疗性抗菌药物使用前分离菌株及药敏情况

目的了解住院患者首次治疗性使用抗菌药物前分离菌株及药敏结果,为医院抗菌药物合理使用提供可靠的微生物依据。方法基于某院抗菌药物临床决策支持系统后台相关节点数据及医院检验信息系统收集的住院患者分离菌株及药敏结果,收集2015—2018年患者入院后首次治疗性使用抗菌药物前临床分离菌株和住院期间送检标本所有临床分离菌株,比较抗菌药物使用前后菌株分布及其耐药情况。结果2015—2018年住院患者所有送检标本共检出非重复菌株69037株(包括细菌及真菌),其中革兰阳性菌17900株(占25.93%),革兰阴性菌44055株(占63.81%),真菌7082株(占10.26%)。住院患者入院后首次治疗性使用抗菌药物前采集标本中检出非重复菌株15017株,其中革兰阳性菌4661株(占31.04%),革兰阴性菌9451株(占62.93%),真菌905株(占6.03%)。治疗用药前标本分离菌居前5位的依次是大肠埃希菌、肺炎克雷伯菌、金黄色葡萄球菌、铜绿假单胞菌和凝固酶阴性葡萄球菌,所有送检标本分离菌居前5位的分别为肺炎克雷伯菌、鲍曼不动杆菌、铜绿假单胞菌、假丝酵母菌属、大肠埃希菌,苛养菌如链球菌属、流感嗜血杆菌、卡他莫拉菌在所有送检标本中占比,较用药前低。临床常见分离菌包括金黄色葡萄球菌、大肠埃希菌、肺炎克雷伯菌、铜绿假单胞菌、鲍曼不动杆菌,抗菌治疗后送检标本分离菌株中耐药率更高。结论研究首次报道大样本治疗性抗菌药物使用前临床送检标本分离的菌株及药敏情况,临床常见分离菌包括金黄色葡萄球菌、大肠埃希菌、肺炎克雷伯菌、铜绿假单胞菌、鲍曼不动杆菌,在抗菌治疗后标本中显示更高的耐药性。

海水循环泵叶轮多参数组合变型设计研究

在某核电站海水循环泵实际运行中,其选型扬程高于实际需要扬程的上限值,针对这一问题,以原模型额定工况点扬程的80%为变型目标,提出了一种基于多参数组合变型的叶轮改造设计方法。首先,在原有模型的基础上,选择叶轮外径、叶片出口安放角、叶片包角和叶轮出口宽度为研究因素,通过基于计算流体力学(CFD)数值模拟的正交试验设计,进行了变型参数分析;然后,总结了各参数与海水循环泵性能之间的关系,进行了变型方案设计,并比较了单因素变型设计与多因素耦合变型设计之间的区别,得到了主要参数的组合变型方案;最后,通过适配其他变型参数,得到了最优变型组合方案,分析了降低扬程和提升效率的流场特点,得到了多参数组合变型对海水循环泵叶轮性能的影响规律。研究结果表明:在额定工况点,变型后的海水循环泵扬程降低了18.57%,效率提升了0.63%,相比仅改变叶轮外径和仅改变叶片出口安放角降低扬程,循环泵的效率分别提高了7.52%和3.27%。

基于WGAN的不均衡太赫兹光谱识别

物质的太赫兹光谱具有唯一性。目前,结合先进的机器学习方法,研究基于规模光谱数据库的太赫兹光谱识别技术已成为太赫兹应用技术领域的重点。考虑到由于实验条件及实验设备的影响,很难收集到多物质均衡光谱数据,而这又是对太赫兹光谱数据进行分类的基础。针对这一问题,提出一种基于WGAN的不均衡太赫兹光谱识别方法。WGAN作为生成数据的一种新方法,将模型达到纳什均衡条件下的生成数据用来补充数据集,使其达到类别均衡。生成数据可以有效映射真实数据分布,通过将生成数据与真实数据混合训练可以提高识别不均衡光谱数据的准确率。采用三种特征谱较为相似的麦芽糖化合物的太赫兹透射光谱数据进行验证,首先利用S-G滤波和三次样条插值法对三种物质的光谱数据进行归一化处理,然后通过构建WGAN模型对三种物质的不均衡太赫兹光谱数据进行扩展,使其达到类别均衡。实验在同一测试集下进行验证,并利用三组对比实验证明WGAN在不均衡数据集处理中的效果。首先利用WGAN生成数据,随着迭代次数的增加,生成数据逐渐符合真实数据分布。实验结果证明,使用WGAN扩展后的数据集训练SVM模型,可以解决模型在测试集上小样本数据(Maltotriose,Malthexaose)偏向大样本数据(Maltoheptaose)的问题。在将WGAN与传统处理不均衡数据集方法FWSVM和COPY对比后发现,三种分类算法在dataset-1数据集上的训练集准确率都能达到90%以上。但是由于模型泛化能力的限制,传统方法在测试集上的效果并不是很理想,而使用WGAN后的测试集准确率却能达到91.54%。在不同不均衡度方面,采用不均衡度为16,81和256的数据集进行验证,其三个测试集上的准确率分别为92.08%,91.54%和90.27%,可满足实际工作中处理不同不均衡度的要求。

Evolution Analysis About Soybean MIR166 Family

MicroRNA（miRNA） is a class of important regulating non-coding small molecular RNA. The gma-MIR166 gene family consists of 21 members and their expression patterns diversify widely. It is important to analyze the evolution of gma-MIR166 gene family in order to understand the evolutionary mechanisms of miRNAs in soybean. In this study, we implemented soybean wide genome block analysis, the molecular phylogeny of gma-MIR166 and block analysis of gma-MIR166 family. The results showed that both chromosome big segmental duplications and tandem duplications were main reasons contributed to the expanding of gmaMIR166 gene family. These findings suggested that gma-MIR166 gene family might originate from one or two ancient miRNA genes. The results of research provided a support for evolutionary study of miRNAs in soybean and related species in Fabaceae.

Identification of candidate genes related to soluble sugar contents in soybean seeds using multiple genetic analyses

Soluble sugar content in seeds is an important quality trait of soybean. In this study, 57 quantitative trait loci(QTLs) related to soluble sugar contents in soybean seeds were collected from databases and published papers. After meta-overview-collinearity integrated analysis to refine QTL intervals, eight consensus QTLs were identified. To further verify the consensus QTLs, a population of chromosome segment substitution lines(CSSLs) was analyzed. Two lines containing fragments covering the regions of consensus QTLs and the recurrent parent were selected: one line showed high soluble sugar contents associated with a consensus QTL fragment, and the other line showed low soluble sugar contents. Transcriptome sequencing was conducted for these two lines at the early, middle, and late stages of seed development, which identified 158, 109 and 329 differentially expressed genes, respectively. Based on the analyses of re-sequencing data of the CSSLs and the consensus QTL region, three candidate genes(Glyma.19 G146800, Glyma.19 G122500, and Glyma.19 G128500) were identified in the genetic fragments introduced from wild soybean. Sequence comparisons between the two CSSL parents SN14 and ZYD00006 revealed a single nucleotide polymorphism(SNP) mutation in the coding sequence of Glyma.19 G122500, causing a nonsynonymous mutation in the amino acid sequence that affected the predicted protein structure. A Kompetitive allele-specific PCR(KASP) marker was developed based on this SNP and used to evaluate the CSSLs. These results lay the foundation for further research to identify genes related to soluble sugar contents in soybean seeds and for future soybean breeding.

Characterization of chromosome segment substitution lines reveals candidate genes associated with the nodule number in soybean

Soybean is one of the most important food crops worldwide.Like other legumes,soybean can form symbiotic relationships with Rhizobium species.Nitrogen fixation of soybean via its symbiosis with Rhizobium is pivotal for sustainable agriculture.Type Ⅲ effectors(T3Es)are essential regulators of the establishment of the symbiosis,and nodule number is a feature of nitrogen-affected nodulation.However,genes encoding T3Es at quantitative trait loci(QTLs)related to nodulation have rarely been identified.Chromosome segment substitution lines(CSSLs)have a common genetic background but only a few loci with heterogeneous genetic information;thus,they are suitable materials for identifying candidate genes at a target locus.In this study,a CSSL population was used to identify the QTLs related to nodule number in soybean.Single nucleotide polymorphism(SNP)markers and candidate genes within the QTLs interval were detected,and it was determined which genes showed differential expression between isolines.Four candidate genes(GmCDPK28,GmNAC1,GmbHLH,and GmERF5)linked to the SNPs were identified as being related to nodule traits and pivotal processes and pathways involved in symbiosis establishment.A candidate gene(GmERF5)encoding a transcription factor that may interact directly with the T3E NopAA was identified.The confirmed CSSLs with important segments and candidate genes identified in this study are valuable resources for further studies on the genetic network and T3Es involved in the signaling pathway that is essential for symbiosis establishment.

GmDRR1,a dirigent protein resistant to Phytophthora sojae in Glycine max (L.) Merr.

Soil-borne pathogen Phytophthora sojae is an oomycete that causes devastating damage to soybean yield. To mine original resistant genes in soybean is an effective and environmentally-friend approach controlling the disease. In this study, soybean proteins were extracted from the first trifoliolates infected by predominant P. sojae race 1 and analyzed by twodimensional gel electrophoresis. Nineteen differently-expressed protein spots were detected, and 10 of them were further applied for Matrix-Assisted Laser Desorption/Ionization Time of Flight Mass Spectrometry Assay. One protein containing a dirigent （DIR） domain was identified and belonged to the DIR-b/d family. Therefore, it was named as GmDRR1 （Glycine max Disease Resistance Response 1）. Then, GmDRR1 gene was pathologically confirmed to be involved in the resistant to P. sojae in soybean. GmDRR1-GFP （green fluorescent protein） fusion proteins localized in the cell membrane. qRTPCR results showed GmDRR1 gene expressed differently in P. sojae resistant- and susceptible-soybean cultivars. By the promoter analysis, we found a haplotype H8 was existing in most resistant soybean varieties, while a haplotype H77 was existing in most susceptible soybean varieties. The H77 haplotype had seven SNPs （C to A, G to C, C to A, T to A, T to C, T to C, and T to A） and two single nucleotide insertions. The results supported that the expression difference of GmDRR1 genes between P. sojae resistant- and susceptible-soybean cultivars might depend on the GmDRR1 promoter SNPs. The results suggested that GmDRR1 was a dirigent protein involved in soybean resistant to P. sojae and paved a novel way for investigation of the molecular regulatory mechanism of the defense response to P. sojae in soybean.

结核分枝杆菌分泌蛋白融合蛋白纳米粒可溶性微针贴片的制备及结核病皮试的应用

结核病(tuberculosis,TB)流行病快速筛查常采用结核菌素纯蛋白衍生物(pure protein derivative,PPD)皮试,但特异性低,不良反应大。ESAT-6和CFP-10是结核分枝杆菌的分泌蛋白,但卡介苗(Bacillus Calmette-Guerin,BCG)中缺失了相关基因。本研究表达纯化了结核分枝杆菌的分泌蛋白ESAT6和CFP10的融合蛋白(ESAT6 and CFP10 fusion protein,EC),并制备成其壳聚糖纳米粒(ESAT6 and CFP10 fusion protein nanoparticals,EC-NPs),进一步制备成透明质酸可溶性微针贴片(microneedle patch,MNP)。MNP-EC-NPs载药量达到每针0.03μg,每片可达到1.92μg,常温下保存期可达6个月,可有效透过表皮。志愿者皮试结果显示MNP-EC-NPs可有效区分BCG接种者,在结核病患者皮肤有效显示阳性反应,无显著不良反应,实验经武汉市肺科医院伦理委员会批准[武肺伦理2022(2)]。本研究建立了一种结核病皮试方法,采用ESAT6-CFP10融合蛋白替代PPD,并采用可溶性微针剂型,提高了结核病皮试诊断的特异性,为结核流行病筛查提供了新的技术方案。