基于粒子群算法的最小二乘支持向量机在红花提取液近红外定量分析中的应用

提出一种基于粒子群算法的最小二乘支持向量机(PSO-LS-SVM)方法,用于建立红花提取过程关键质控指标的定量分析模型.近红外光谱数据经波段选择、预处理和主成分分析(降维)后,利用粒子群优化(PSO)算法对最小二乘支持向量机算法中的参数进行优化,然后使用最优参数建立固含量和羟基红花黄色素A(HSYA)浓度的定量校正模型.将校正结果与偏最小二乘法回归(PLSR)和BP神经网络(BP-ANN)比较,并将所建的3个模型用于红花提取过程未知样本的预测.结果表明,BP-ANN校正结果优于PSO-LS-SVM和PLSR,但是对验证集和未知样品集的预测能力较差,而PSO-LS-SVM和PLSR模型的校正、验证结果相近,相关系数均大于0.987,RMSEC和RMSEP值相近且小于0.074,RPD值均大于6.26,RSEP均小于5.70％.对于未知样品集,pSO-LS-SVM模型的RPD值大于8.06,RMSEP和RSEP值分别小于0.07％和5.84％,较BP-ANN和PLSR模型更低.本研究所建立的PSO-LS-SVM模型表现出较好的模型稳定性和预测精度,具有一定的实践意义和应用价值,可推广用于红花提取过程的近红外光谱定量分析.     

基于平均影响值的反向传播神经网络方法用于提高密度泛函理论计算Y—NO键均裂能精度

将基于平均影响值(Mean impact value,MIV)的反向传播神经网络(Back propagation neural netowrk,BPNN)(MIV-BPNN)方法用于提高密度泛函理论(Density functional theory,DFT)计算Y—NO(Y=N,S,O及C)键均裂能的精度.利用量子化学计算和MIV-BPNN联合方法计算92个含Y—NO键的有机分子体系的均裂能.结果表明,相对于单一的密度泛函理论B3LYP/6-31G(d)方法,利用全参数下的BPNN方法计算92个有机分子均裂能的均方根误差从22.25 kJ/mol减少到1.84 kJ/mol,而MIV-BPNN方法使均方根误差减少到1.36 kJ/mol,可见B3LYP/6-31G(d)和MIV-BPNN联合方法可以提高均裂能的量子化学计算精度,并可预测实验上无法获取的均裂能值.     

基于启发式方法和支持向量机方法预测有机物的热导率

构建147个有机物分子结构与其热导率值之间的定量结构-性质关系(QSPR)模型, 探讨影响有机物热导率的结构因素. 以147个化合物作为样本集, 随机选择118个作为训练集, 29个作为测试集. 应用CODESSA软件计算了组成、拓扑、几何、静电和量子化学等描述符, 通过启发式方法(HM)筛选得到5个结构参数并建立线性回归模型; 用所选5个结构参数作为支持向量机(SVM)的输入, 建立非线性的支持向量机回归模型. 预测结果表明: 支持向量机回归模型的性能(复相关系数R²=0.9240)虽略低于启发式回归模型的性能(R²=0.9267), 但是支持向量机方法预测性能(R²=0.9682)高于启发式方法的预测性能(R²=0.9574), 对于QSPR模型来说, 预测性能更重要. 因此, 总体来说支持向量机方法优于启发式方法. 支持向量机方法和启发式方法的提出为工程上提供了一种根据分子结构预测有机物热导率的新方法.     

中药材三七提取液近红外光谱的支持向量机回归校正方法

提出近红外光谱的支持向量机回归校正建模方法.以中药材三七渗漉提取液为实际分析对象,对其近红外光谱数据进行预处理和主成分分析后,用支持向量机回归算法建立人参皂苷R_g1,R_b1和R_d以及三七总皂苷的近红外光谱校正模型.以R_g1,R_b1和R_d的HPLC测定值及三七总皂苷的比色法测定值为参照,将本文方法与偏最小二乘回归和径向基神经网络建模方法相比较,结果表明,本文所建模型的预测准确性优于后两者,可推广应用于中药提取过程的近红外光谱分析.     

基于机器学习方法的丙型肝炎病毒聚合酶NS5B非核苷抑制剂的定量构效关系研究

对89 个苯并异噻唑和苯并噻嗪类丙型肝炎病毒(HCV) NS5B聚合酶非核苷抑制剂进行了定量构效关系(QSAR)研究. 采用遗传算法组合偏最小二乘(GA-PLS)和线性逐步回归分析(LSRA)两种特征选择方法选择最优描述符子集, 然后建立多元线性回归和偏最小二乘线性回归模型. 并首次尝试使用遗传算法耦合支持向量机方法(GA-SVM)对两种特征选择方法所选的描述符子集分别建立非线性支持向量机回归模型. 三种机器学习方法所建模型均得到比较满意的预测效果. 采用LSRA所选的6 个描述符建立的三个QSAR模型对于测试集的相关系数为0.958-0.962, GA-SVM法给出最好的预测精度(0.962). 采用GA-PLS所选的7个描述符建立的三个QSAR模型对于测试集的相关系数为0.918-0.960, 偏最小二乘回归模型的结果最好(0.960). 本工作提供了一种有效的方法来预测丙型肝炎病毒抑制剂的生物活性, 该方法也可以扩展到其他类似的定量构效关系研究领域.     

基于最小二乘支持向量机的色谱指纹图谱预测银杏叶总抗氧化活性

在色谱图基线校正和色谱峰匹配基础上,提出以40个银杏叶提取物HPLC指纹图谱的色谱图轮廓作为输入,相应的提取物总抗氧化活性作为输出,建立最小二乘支持向量机回归模型,并对包含10个样本的测试集进行了预测.最小二乘支持向量机的测试集预测误差均方根(RMSEP)为0.0230,预测结果优于目前普遍使用的误差反向传播神经网络和偏最小二乘回归.与采用色谱峰面积为分析变量的模型预测结果比较表明:采用消除干扰后的色谱图全谱轮廓保留了样本的全部信息,预测结果更好     

预测毛细管区带电泳有效淌度的支持向量回归建模方法

提出预测毛细管电泳迁移行为的支持向量回归建模方法。以核苷为实际研究对象,利用正交试验获得的数据,结合二标记物技术,用支持向量回归算法建立毛细管区带电泳的柱温、电压、缓冲液浓度和pH值与3种核苷的有效淌度之间的相关模型。将其与偏最小二乘回归和人工神经网络方法相比较,结果表明所建模型的预测准确性优于后两者,适宜用于毛细管电泳迁移行为的预测。     

基于最小二乘支持向量机的油页岩含油率近红外光谱分析

为了提高油页岩含油率近红外光谱分析建模的预测精度和稳定性,开展了基于最小二乘支持向量机(LS-SVM)建模方法的对比研究.采用主成分-马氏距离(PCA-MD)和基于蒙特卡洛采样(MCS)2种方法进行了奇异样本的检测,采用径向基核函数的LS-SVM、偏最小二乘(PLS)和反向传播神经网络(BPANN)3种方法进行建模方法对比.结果表明,对于64个油页岩岩芯样本,与PCA-MD方法相比,采用MCS方法剔除奇异样本后所建PLS模型的预测精度提高了28%.对于MCS方法剔除奇异样本后的58个样品,采用KennardStone法划分了44个样品的校正集和14个样品的预测集,采用2阶导数和标准化预处理方法,建立了100个LS-SVM的校正模型,模型的预测决定系数R2平均值达到0.90以上,高于PLS和BPANN模型的对应值;且R2的变化量(0.02)小于BPANN模型的对应值(0.32).因此,MCS奇异样本检测结合LS-SVM方法可提高油页岩含油率样本建模的精度和稳定性.     

支持向量机与KStar模型预测细胞色素P450酶催化的氧脱烃反应

分别以支持向量机(SVM)和KStar方法为基础, 构建了代谢产物的分子形状判别和代谢反应位点判别的嵌套预测模型. 分子形状判别模型是以272个分子为研究对象, 计算了包括分子拓扑、二维自相关、几何结构等在内的1280个分子描述符, 考查了支持向量机、决策树、贝叶斯网络、k最近邻这四种机器学习方法建立分类预测模型的准确性. 结果表明, 支持向量机优于其他方法, 此模型可用于预测分子能否被细胞色素P450酶催化发生氧脱烃反应. 代谢反应位点判别模型以538个氧脱烃反应代谢位点为研究对象, 计算了表征原子能量、价态、电荷等26个量子化学特征, 比较了决策树、贝叶斯网络、KStar、人工神经网络建模的准确率. 结果显示, KStar模型的准确率、敏感性、专一性均在90%以上, 对分子形状判别模型筛选出的分子, 此模型能较好地判断出哪个C―O键发生断裂. 本文以15个代谢反应明确的中药分子为验证集, 验证模型准确性, 研究结果表明基于SVM和KStar的嵌套预测模型具有一定的准确性, 有助于开展中药分子氧脱烃代谢产物的预测研究.     

基于支持向量机的汽油族组成近红外光谱分析方法研究

提出一种基于最小二乘支持向量机(LS-SVM)的汽油族组成近红外光谱分析方法。采用国家标准方法(GB 11132-1989液体石油产品烃类测定法——荧光指示剂吸附法)测定了重庆地区销售的汽油族组成,并采用主成分分析-最小二乘支持向量机建立汽油族组成的预测模型。预测模型对汽油中芳烃和烯烃含量的RMSEC分别为0.2090和0.2142。实验结果表明所建模型具有计算量小,预测准确、可靠,而且操作简单、维护费及测试费用低等特点。     

Improving the accuracy of density-functional theory calculation: the genetic algorithm and neural network approach

The combination of genetic algorithm and neural network approach (GANN) has been developed to improve the calculation accuracy of density functional theory. As a demonstration, this combined quantum mechanical calculation and GANN correction approach has been applied to evaluate the optical absorption energies of 150 organic molecules. The neural network approach reduces the root-mean-square (rms) deviation of the calculated absorption energies of 150 organic molecules from 0.47 to 0.22 eV for the TDDFTB3LYP6-31G(d) calculation, and the newly developed GANN correction approach reduces the rms deviation to 0.16 eV.     

Electron impact ionisation and UV absorption study of α- and β-pinene

We have performed a coordinated set of experiments to measure the electron impact ionisation and UV photoabsorption cross sections of α- and β-pinene. The adiabatic ionisation energies of α- and β-pinene were derived from experiment and found to be 8.3 and 8.6 eV which compared well with high-level quantum chemical calculations (G3MP2) yielding values of 8.29 and 8.41 eV. Additionally, vertical ionisation energies of 8.62 and 8.96 eV were calculated using an OVGF method. UV photoabsorption cross sections were measured using a high-resolution synchrotron radiation source and electronic states interpreted on the basis of the TD quantum chemical methods.     

一种对大体系局部进行高精度量子化学计算的新方法

提出一种可以对大体系的局部区域进行高精度量子化学计算的新方法.将体系划分为活性区和环境区,采用比较粗略的算法获得大体系的密度矩阵,通过基组的变换将密度矩阵分解为对应于活性区与环境区的子矩阵的直和,活性区和环境区分别拥有整数个电子,成为相对独立的化学键饱和的子体系.然后冻结环境区密度矩阵,同时考虑环境区在活性区所产生势场的作用,对活性区进行高精度的量子化学计算.对一些分子进行计算,得到的电荷分布、键长、键断裂能、电离势与电负性等与对大体系整体进行高精度计算得到的结果相当一致,而计算量减少很多,表明提出的方法可行,并有应用价值.     

A comparative study of semiempirical, ab initio, and DFT methods in evaluating metal-ligand bond strength, proton affinity, and interactions between first and second shell ligands in Zn-biomimetic complexes

Although theoretical methods are now available which give very accurate results, often comparable to the experimental ones, modeling chemical or biological interesting systems often requires less demanding and less accurate theoretical methods, mainly due to computer limitations. Therefore, it is crucial to know the precision of such less reliable methods for relevant models and data. This has been done in this work for small zinc-active site models including O- (H(2)O and OH(-)) and N-donor (NH(3) and imidazole) ligands. Calculations using a number of quantum mechanical methods were carried out to determine their precision for geometries, coordination number relative stability, metal-ligand bond strengths, proton affinities, and interaction energies between first and second shell ligands. We have found that obtaining chemical accuracy can be as straightforward as HF geometry optimization with a double-zeta plus polarization basis followed by a B3LYP energy calculation with a triple-zeta quality basis set including diffuse and polarization functions. The use of levels as low as PM3 geometry optimization followed by a B3LYP single-point energy calculation with a double-zeta quality basis including polarization functions already yields useful trends in bond length, proton affinities or bond dissociation energies, provided that appropriate caution is taken with the optimized structures. The reliability of these levels of calculation has been successfully demonstrated for real biomimetic cases.     

Frozen-density embedding-based many-body expansions

Fragmentation methods allow for the accurate quantum chemical (QC) treatment of large molecular clusters and materials. Here we explore the combination of two complementary approaches to the development of such fragmentation methods: the many-body expansion (MBE) on the one hand, and subsystem density-functional theory (DFT) or frozen-density embedding (FDE) theory on the other hand. First, we assess potential benefits of using FDE to account for the environment in the subsystem calculations performed within the MBE. Second, we use subsystem DFT to derive a density-based MBE, in which a many-body expansion of the electron density is used to calculate the system's total energy. This provides a correction to the energies calculated with a conventional energy-based MBE that depends only on the subsystem's electron densities. For the test case of clusters of water and of aspirin, we show that such a density-based MBE converges faster than the conventional energy-based MBE. For our test cases, truncation errors in the interaction energies are below chemical accuracy already with a two-body expansion. The density-based MBE thus provides a promising avenue for accurate QC calculation of molecular clusters and materials.     

A global potential energy surface and time‐dependent quantum wave packet calculation of Au + H2 reaction

A global potential energy surface (PES) corresponding to the ground state of AuH₂ system has been constructed based on 22 853 ab initio energies calculated by the multireference configuration interaction method with a Davidson correction. The neural network method is used to fit the PES, and the root mean square error is only 1.87 meV. The topographical features of the novel global PES are compared with previous PES which is constructed by Zanchet et al. (Zanchet PES). The global minimum energy reaction paths on the two PESs both have a well and a barrier. Relative to the Au + H₂ reactants, the energy of well is 0.316 eV on the new PES, which is 0.421 eV deeper than Zanchet PES. The calculation of Au(²S) + H₂(X¹Σ_g⁺) → AuH(X¹Σ⁺) + H(²S) dynamical reaction is carried out on new PES, by the time‐dependent quantum wave packet method (TDWP) with second order split operator. The reaction probabilities, integral cross‐sections (ICSs) and differential cross‐sections are obtained from the dynamics calculation. The threshold in the reaction is about 1.46 eV, which is 0.07 eV smaller than Zanchet PES due to the different endothermic energies on the two PESs. At low collision energy (<2.3 eV), the total ICS is larger than the result obtained on Zanchet PES, which can be attributed to the difference of the wells and endothermic energies.     

Ladder oligo(m-aniline)s: derivatives of azaacenes with cross-conjugated π-systems

We describe the synthesis and electronic properties of ladder oligomers of poly(m-aniline) that may be considered as derivatives of azaacenes with cross-conjugated π-systems. Syntheses of ladder oligo(m-aniline)s with 9 and 13 collinearly fused six-membered rings employed Pd-catalyzed aminations and Friedel-Crafts-based ring closures. Structures were confirmed by either X-ray crystallography or correlations between DFT-computed and experimental spectroscopic data such as (1)H, (13)C, and (15)N NMR chemical shifts and electronic absorption spectra. All compounds have planar "azaacene" moieties. The experimental band gaps E(g) ≈ 3.5-3.65 eV, determined by the UV-vis absorption onsets, were in agreement with the TD-DFT-computed vertical excitation energies to the S(1) state. Fluorescence quantum yields of up to 20% were found. Electrochemically estimated HOMO energies of -4.8 eV suggested propensity for a facile one-electron oxidation and just sufficient environmental stability toward oxygen (O(2)). For two oligomers with "tetraazanonacene" moieties, potentials of E(4+/3+) ≈ 1.6-1.7 V vs SCE were determined for four-electron oxidation to the corresponding tetraradical tetracations.     

Barrier Height Effect on Cl+H2(D2) Reaction

Three-dimensional time-dependent quantum wave packet calculation was performed to study the reaction dynamics of Cl+H2(D2) on two potential energy surfaces (CW PESs). The first CW PES is with spin-orbit correction; the second is without spin-orbit correction. The integral cross-section and reaction probability as a function of collision energy are calculated in the collision energy range of 0.1 eV to 1.4 eV. For reaction of Cl with D2, the reaction section with spin-orbit correction has a shift toward the high energy because the barrier height increases. As for the reaction of Cl with H2 at low collision energy, it is more reactive on the PES with spin-orbit correction than on the low barrier height PES without spin-orbit correction, due to the tunnel effect for the reaction of the Cl with H2. When the collision energy is higher than 0.7 eV, the reactivity on the low barrier height PES is larger than that on the high barrier height PES. It is believed that the barrier height plays a very important role in the reactivity of Cl with (H2, D2). For the Cl+H2 reaction the barrier width is also very important because of the tunneling effect.     

Low temperature synthesis of cubic phase zinc sulfide quantum dots

In this study, we report on a new method for the synthesis of ZnS quantum dots (QDs). The synthesis was carried out at low temperature by a chemical reaction between zinc ions and freshly reduced sulfide ions in ethanol as reaction medium. Zinc chloride and elemental sulfur were used as zinc and sulfur sources, respectively and hydrazine hydrate was used as a strong reducing agent to convert elemental sulfur (S₈) into highly reactive sulfide ions (S²⁻) which react spontaneously with zinc ions. This facile, less toxic, inexpensive route has a high yield for the synthesis of high quality metal sulfide QDs. Transmission electron microscopy (TEM) image analysis and selected area electron diffraction (SAED) reveal that ZnS QDs are less than 3 nm in diameter and are of cubic crystalline phase. The UV-Vis absorption spectrum shows an absorption peak at 253 nm corresponding to a band gap of 4.9 eV, which is high when compared to the bulk value of 3.68 eV revealing strong quantum confinement. PL emission transitions are observed at 314 nm and 439 nm and related to point defects in ZnS QDs.