引入模拟辅助样本提高BP网络的泛化能力

以2－吲哚醇在20种不同参数固定液上的保留值、19种不同物质在同一固定液上的保留值以及脂肪胺的电谱保留值分别作为网络的训练样本和检验样本，建立了多元线性回归（LR）模型和BP网络模型，并基于LR模型运用随机搜索最优化方法，产生模拟辅助样本并将其引入BP网络训练样本集。预测结果表明，该方法的使用提高了BP网络的泛化能力，对于残缺样本问题的预测研究，提供了一种有效的方法，与线性回归模型及原BP网络模型相比，预测精度有了明显的改善。     

人工神经网络用于消化道癌症初级诊断研究

测定了正常人及消化道癌症病人头发样品１１种元素含量,然后将人工神经网络（ＡＮＮ）用于正常人与癌症患者的分类预测,并用独立预测样本作了检验,预报识别率达１００％。讨论了网络参数的选择和发样中微量元素与癌症的关系。结果表明该方法可作为消化道癌症初级诊断的一种辅助手段。     

苹果产地差异对可溶性固形物近红外光谱检测模型影响的研究

为更好地利用近红外光谱预测苹果可溶性固形物含量,减少产地差异对近红外光谱检测模型的影响,以4种不同产地的富士苹果为研究对象,采用基于x-y共生距离的样本划分方法分别对不同产地的苹果选取代表性样本作为校正集,利用偏最小二乘算法,建立和比较单一产地和混合产地下的苹果可溶性固形物近红外光谱检测模型,并结合竞争性自适应重加权算法(CARS)和连续投影算法(SPA)对苹果可溶性固形物的建模变量进行筛选。相比单一产地和其它混合产地模型,混合所有4种苹果产地的校正集样本建立的模型取得了最好的预测结果,另外,结合CARS-SPA筛选的16个特征波长,模型得到了进一步简化,其预测相关系数和预测均方根误差分别为0.978和0.441°Brix。结果表明,利用多个产地的苹果样本建立的混合模型,结合有效特征波长,可提高对苹果可溶性固形物含量的预测精度,减小产地差异对可溶性固形物近红外光谱检测的影响。     

模式识别辅助镧镍电极的设计

应用模式识别和人工神经网络方法将已知样本按目标（稀土系镍氢电池容量和循环寿命）分类识别建模，在此基础地若干样本，通过人工神经网络预报，电池容量和循环寿命都达到设计要求。     

基于平行因子分析法提高近红外无创血糖校正模型稳健性的研究

采用近红外光谱进行无创血糖检测时, 样品背景变动造成的预测集样本与校正集样本量测体系不一致的问题是导致预测精度低的原因之一. 提出一种将母体背景作为变量引入回归建模中, 结合各个母体背景下的样本光谱信息构建三维光谱矩阵以提高校正模型稳健性的分析方法. 将平行因子分析(PARAFAC)与多元线性回归(MLR)相结合, 对人体三层皮肤模型的蒙特卡罗模拟实验和葡萄糖水溶液及其混合物的离体实验进行了验证. 实验结果表明, 与传统的单一母体背景所建立的偏最小二乘模型相比, 将母体背景作为建模元素采用PARAFAC-MLR法所建立的校正模型具有更好的预测能力和稳健性.     

短波近红外光谱技术对葡萄酒中总糖含量快速测定的研究

采用短波近红外光谱技术结合偏最小二乘法(PLS),建立了葡萄酒中总糖含量的定量分析数学模型,讨论了光谱预处理方法和主成分数对PLS模型预报精度的影响.应用所建模型对预测集样本中总糖含量进行预报,结果令人满意.该方法方便快捷,并且具有较高的预报精度,可以用于葡萄酒中总糖含量的快速测定.     

基于高光谱技术的培养基上细菌菌落分类方法研究

利用高光谱技术对培养基上细菌(大肠杆菌、李斯特菌和金黄色葡萄球菌)菌落进行快速识别和分类。采集琼脂培养基上细菌菌落的高光谱反射图像(390~1040 nm),在对波段差图像进行大津阈值分割的基础上自动提取细菌菌落光谱,并建立细菌分类检测的全波长和简化偏最小二乘判别( PLS-DA)模型。全波长模型对预测集样本的分类准确率和置信预测分类准确率分别为100%和95.9%。此外,利用竞争性自适应重加权算法( CARS)、遗传算法( GA)和最小角回归算法( LARS-Lasso)进行波长优选并建立对应简化模型。其中,CARS简化模型在精度、稳定性及分类准确率方面均优于GA和LARS-Lasso简化模型,其对预测集样本的分类准确率和置信预测分类准确率分别达到了100%和98.0%。研究表明,高光谱是一种细菌菌落高精度、快速、无损识别检测的有效方法。简化模型中优选的波长可以为开发低成本检测仪器提供理论依据。     

基于高光谱图像的生菜叶片氮素含量预测模型研究

为了便于更经济合理地为作物施肥,建立一种无损检测作物氮营养元素的高光谱图像模型。本实验以生菜为研究对象,无土栽培各氮素水平的生菜叶样本,在莲座期,采集生菜叶片样本的高光谱图像(390~1050 nm),同时采用凯氏定氮法测定对应生菜叶片样本的全氮含量。通过ENVI软件提取出生菜叶片中感兴趣区域的平均光谱作为该样本原始光谱信息,分别使用平滑处理(Smoothing)、多元散射矫正(MSC)、标准正态变量变换结合去趋势(SNV detrending)、一阶导数法(First derivative)、二阶导数法(Second derivative)、正交信号矫正(OSC)等预处理方法对样本原始光谱进行处理,然后利用偏最小二乘回归法(Partial least squares regression,PLSR)分别建立样本全波段光谱信息与氮含量的关系模型,研究各预处理方法对氮含量模型的影响,结果表明,使用OSC预处理的模型效果最好。为了简化模型,根据OSC预处理光谱后的模型的PLSR回归系数优选出敏感波长,利用训练集中样本的敏感波长光谱信息与氮含量数据重新构建PLSR回归模型,并利用测试集样本进行测试试验。结果表明,该模型得到校正集和预测集的决定系数(R2p)分别为0.89,0.81;均方根误差RMSEC,RMSEP分别为0.33,0.45。该回归模型大大降低了自变量个数,简化了模型,并且取得了较优的效果,这为生菜氮素含量预测提供了一种新的快速有效方法。     

叶酸及其相关代谢物在神经管畸形风险评估和预测中的作用比较研究

以叶酸及其相关代谢物为研究对象,结合定量分析方法以及多种统计学方法,比较它们在神经管畸形发病风险评估和预测中的作用。采用前期建立的HPLC-MS/MS方法测定了121例病例和118例对照母体血清中的代谢物水平,通过双样本成对t检验筛选差异代谢物(p<0.05),然后利用Logistic回归和ROC分析计算这些代谢物的优势比以及预测灵敏度及特异性、阳性和阴性预测率。结果表明,4种差异代谢物中,5-甲基四氢叶酸和同型半胱氨酸的优势比、阳性和阴性预测率均高于叶酸。本方法适用于疾病标志物的筛选和评价。     

选择岭回归参数k值的改进方法及其在农药定量构效关系中的应用

提出改进的预报相对误差法选择岭回归参数ｋ，以平均预报相对误差替代预报残差平方和，并抑制过拟合。该方法应用于苯乙酰胺类除草农药定量构效关系的二次建模，效果良好，预报精度优于残差平方和方法。     

一种基于LASSO回归模型的彩色相机颜色校正方法

彩色相机的颜色校正是实现成像色彩一致性的必要保障手段。传统的相机颜色校正中,对测量数据多采用多项式回归分析来确定颜色定标系数,存在着精度不高的缺点,因此,本文对测量数据提出了基于LASSO的高阶多项式回归拟合方法,利用LASSO压缩系数的特点,在保证计算复杂度的前提下,有效提高了回归模型的校正精度。在D65标准光源下对ColorChecker 24色卡进行了实际成像实验,并用CIELAB色差公式表征了校正效果,实验结果表明,新方法的校正效果明显优于传统的线性回归、二次多项式回归方法,平均色差指标可以达到5个CIELAB色差单位。     

QSAR modeling based on the bias/variance compromise: a harmonious

Modeling quantitative structure–activity relationships (QSAR) is considered with an emphasis on prediction. An abundance of methods are available to develop such models. Using a harmonious approach that balances the bias and variance of predictions, the best calibration models are identified relative to the bias and variance criteria used. Criteria utilized to determine the adequacy of models are the root mean square error of calibration (RMSEC) and validation (RMSEV), respective R ² values, and the norm of the regression vector. QSAR data from the literature are used to demonstrate concepts. For these data sets and criteria used, it is suggested that models obtained by ridge regression (RR) are more harmonious and parsimonious than models obtained by partial least squares (PLS) and principal component regression (PCR) when the data is mean-centered. The most harmonious RR models have the best bias/variance tradeoff reflected by the smallest RMSEC, RMSEV, and regression vector norms and the largest calibration and validation R ² values. The most parsimonious RR models have the smallest effective rank.     

Regression models based on new local strategies for near infrared spectroscopic data

In this work, a comparative study of two novel algorithms to perform sample selection in local regression based on Partial Least Squares Regression (PLS) is presented. These methodologies were applied for Near Infrared Spectroscopy (NIRS) quantification of five major constituents in corn seeds and are compared and contrasted with global PLS calibrations. Validation results show a significant improvement in the prediction quality when local models implemented by the proposed algorithms are applied to large data bases.     

Impact of calibrator concentrations and their distribution on accuracy of quadratic regression for liquid chromatography–mass spectrometry bioanalysis

Despite the common use of quadratic regression in LC–MS bioanalysis, how calibrator concentrations should be determined is still vague. Both the number and concentrations of calibrators are usually selected arbitrarily to each one's preference. The purposes of this research were to evaluate the impact of calibrator concentrations and to find new approaches with improved accuracy and reduced cost for LC–MS bioanalysis. It was found for the first time that the lower and upper limits of quantitation plus their geometric mean are the three critical concentrations for quadratic regression. When different concentration ranges, different response precisions, and various degrees of downward quadratic responses were simulated, the best accuracy was obtained by including these critical concentrations and using fewer calibrator concentrations with more replicates per concentration, instead of using more calibrator concentrations in duplicate. In many cases, when the aforementioned three concentrations are used, as few as two replicates per concentration are enough for routine use and up to 20% of time and cost can be saved. Furthermore, downward quadratic response should be eliminated or reduced as much as possible and upper limit quality control must be included in each batch to monitor the accuracy at the high concentration end. The retrospective data analysis of published experimental results corroborates the aforementioned findings. Finally, the typical “concerns” and potential applications of the new quadratic regression approaches are discussed.     

Combining synchronous fluorescence spectroscopy with multivariate methods for the analysis of petrol-kerosene mixtures

Synchronous fluorescence spectroscopy (SFS) is a rapid, sensitive and nondestructive method suitable for the analysis of multifluorophoric mixtures. The present study demonstrates the use of SFS and multivariate methods for the analysis of petroleum products which is a complex mixture of multiple fluorophores. Two multivariate techniques principal component regression (PCR) and partial least square regression (PLSR) have been successfully applied for the classification of petrol-kerosene mixtures. Calibration models were constructed using 35 samples and their validation was carried out with varying composition of petrol and kerosene in the calibration range. The results showed that the method could be used for the estimation of kerosene in kerosene-mixed petrol. The model was found to be sensitive, detecting even 1% contamination of kerosene in petrol.     

Characterization of Chickpea Flour by Near Infrared Spectroscopy and Chemometrics

Near infrared (NIR) spectrometry was used for the rapid characterization of quality parameters in desi chickpea flour (besan). Partial least square regression, principal component regression (PCR), interval partial least squares (iPLS), and synergy interval partial least squares (siPLS) were used to determine the protein, carbohydrate, fat, and moisture concentrations of besan. Spectra were collected in reflectance mode using a lab-built predispersive filter-based instrument from 700 to 2500?nm. The quality parameters were also determined by standard methods. The root mean square error (RMSE) for the calibration and validation sets was used to evaluate the performance of the models. The correlation coefficients for moisture, fat, protein, and carbohydrates in chickpea flour exceeded 0.96 using PLS and PCR models using the full spectral range. Wavelengths from 2100 to 2345?nm had the lowest RMSE for quality parameters by iPLS. The error was further decreased by 0.41, 0.1, and 1.1% for carbohydrates, fats, and proteins by siPLS. The NIR spectral regions yielding the lowest RMSE of prediction were 1620–2345?nm for carbohydrates, 1180–1590?nm and 1860–2094?nm for fat, and 1700–2345?nm for proteins. The study shows that chickpea flour quality parameters were accurately determined using the optimized wavelengths.     

Principal component multinomial regression and spectrometry to predict soil texture

The mineral particles are classified in different textural classes according to their size. Reflectance spectrometry and spectra can be valid instruments to classify the soils according to their texture. This is possible using different statistical methods, for example, discriminant analysis. However, other multivariate methods, like multinomial logistic regression, can be used, but the presence of multicollinearity among explicative variables could affect the estimation of the parameters. The solution proposed to remedy this problem is an alternative way to apply the multinomial logit model. To evaluate its performances, we compare the results with both classical multinomial logit and discriminant analysis ones. Copyright © 2015 John Wiley & Sons, Ltd.     

紫杉烷二萜类化合物的定量构效关系研究

紫杉醇是从紫杉或红豆杉树中提取的一种天然抗癌原料药,具有独特的抗癌机理。由于紫杉醇的种种限制,开发具有更高抗癌活性的类紫杉醇药物具有广阔的前景。紫杉烷二萜是以紫杉醇为母体,通过对其结构的不断修饰得到的一些二代紫杉醇类化合物。本文选用30个结构多样的紫杉烷二帖类化合物作为数据集,随机选取其中24个作为训练集,其它分子作为检验集,采用多元线性回归法(MLR)及主成分回归分析法(PCA)对每个化合物的195个分子参数进行回归分析,分别建立了定量构效关系的最优预测模型;并用检验集检验了所建模型的预测能力。结果表明,多元线性回归法所建模型与主成分回归法所建模型相对比,发现逐步筛选法为最优建模方法。该方法所建模型统计结果良好(R=0.782,SEE=0.202),应用于检验集时结果也比较令人满意(R=0.764,SEP=0.114),模型表现出较强的可靠性和预测性。模型的建立和主要影响因素的确定有助于指导新型紫杉醇类似物药物的筛选和研发。     

A new strategy to improve the predictive ability of the local lazy regression and its application to the QSAR study of melanin‐concentrating hormone receptor 1 antagonists

In the quantitative structure‐activity relationship (QSAR) study, local lazy regression (LLR) can predict the activity of a query molecule by using the information of its local neighborhood without need to produce QSAR models a priori. When a prediction is required for a query compound, a set of local models including different number of nearest neighbors are identified. The leave‐one‐out cross‐validation (LOO‐CV) procedure is usually used to assess the prediction ability of each model, and the model giving the lowest LOO‐CV error or highest LOO‐CV correlation coefficient is chosen as the best model. However, it has been proved that the good statistical value from LOO cross‐validation appears to be the necessary, but not the sufficient condition for the model to have a high predictive power. In this work, a new strategy is proposed to improve the predictive ability of LLR models and to access the accuracy of a query prediction. The bandwidth of k neighbor value for LLR is optimized by considering the predictive ability of local models using an external validation set. This approach was applied to the QSAR study of a series of thienopyrimidinone antagonists of melanin‐concentrating hormone receptor 1. The obtained results from the new strategy shows evident improvement compared with the commonly used LOO‐CV LLR methods and the traditional global linear model. © 2009 Wiley Periodicals, Inc. J Comput Chem, 2010     

小波潜变量回归和广义回归神经网络同时测定三组分混合物

本文采用小波潜变量回归（WLVR）方法，同时测定重叠的光谱信号。结合小波阈值法和主组分分析（PCA）改进除噪质量。八个误差判据用于推断因子数目。潜变量由小波处理过的信号投影到正交基矢量而获得。广义回归神经网络（GRNN）被应用于多组分同时测定。依据算法原理编制了三个程序（PWMRA、PWLVR和PGRNN）执行有关计算。三个方法（WLVR、LVR（潜变量回归）和GRNN）同时测定三组分混合物，获得满意的结果。