BP人工神经网络用于芳香族化合物结构参数和大鼠LD_(50)构效关系研究

对结构参数采用主成分变换 ,再利用 BP人工神经网络 ,采用 L M算法作为迭代方法训练网络 ,预测检验集化合物的LD5 0 。结果显示 ,BP人工神经网络可以用于定量毒性构效关系研究 ,含隐层的 BP人工神经网络拟合能力明显优于传统方法 ,消除过度拟合后的多层 BP网络预测能力也好于传统方法 ,可以用于预测。     

人工神经网络在半夏泻心汤配伍建模中的应用

目的 :应用BP人工神经网络 ,建立半夏泻心汤不同配伍与胃蛋白酶间的非线性映射模型。方法 :应用均匀设计表给出半夏泻心汤中药物及生姜共 8味药物不同配伍组合 ,共形成 2 4组 ,采用安宋氏法测定不同组别对正常大鼠胃蛋白酶活性的影响 ,应用MATLAB 6 .5进行编程 ,选用BP人工神经网络拟合实验数据 ,其中 2 1组作为学习样本 ,建立模型 ,另外 3组作为未学习样本 ,验证模型的预测能力。结果 :通过对 2 1组实验数据的学习 ,建立了拓扑结构为 8 10 1的BP网络模型 ,所建模型可以很好的拟合学习过的样本 ,并且可以很好地预测未学习过的样本 ,预测值和实际值之间的相关性系数r=0 .94 33。结论 :BP神经网络可以很好的拟合复方配伍中复杂的非线性关系 ,可以应用于复方配伍研究的建模。     

基于BP神经网络的半夏、生姜、甘草三泻心汤配伍研究

目的:探讨不同BP(BackPropagation,BP)算法的人工神经网络在半夏泻心汤、生姜泻心汤、甘草泻心汤配伍中的应用,并应用所建模型探讨三复方药味与剂量的配伍规律。方法:应用均匀设计对药味及剂量进行分组,测定不同组别对正常大鼠胃粘液的影响。应用MATLAB6.5进行编程,选用BP神经网络来拟合实验数据,比较831、881、8121三种拓扑结构、不同BP算法对网络模型拟合效果的影响,建立基于BP神经网络的三方对胃粘液含量影响的预测模型。结果:拓扑结构为881、算法为改进BP算法的神经网络模型可以很好的拟合学习过的样本,并对未学习过的样本有较好的预测能力,其中采用动量法和学习速率自适应调整两种策略相结合的改良BP算法的网络拟合预测效果最佳。应用模型分析可以看出,每种药物剂量变化及不同药物组合对胃粘液分泌的影响不尽相同,如辛开组合具有促进胃粘液分泌的作用,苦降组合、甘补组合具有抑制胃粘液分泌的作用。结论:以半夏、甘草、生姜泻心汤为研究模板,提出的复方类方配伍规律研究模式:“优化拆方实验设计-人工智能数据挖掘-复方类方知识发现”,将为复杂复方的研究提供借鉴。     

人工神经网络预测7种药物的毛细管电泳迁移时间

目的建立毛细管电泳迁移时间的人工神经网络的预测方法。方法运用人工神经网络 ,通过毛细管区带电泳 (CZE)的实验电压 (V)和缓冲溶液的离子强度 (I)对维生素B1等 7种药物的迁移时间 (tmig)进行预测 ,网络采用三层结构即输入层 隐含层 输出层 ( 2 4 1型 ) ,权值修正采用误差反向传播算法 ,每种药物的样本数为 5 0 ,以“留一法”预测其迁移时间 ,网络学习次数为 1 0 0 0 0。结果当电压在 6～ 2 6kV以及硼砂溶液的离子强度为 1 0～ 1 0 0mmol/L时 ,网络预测的相对误差绝对值小于 1 2 %的概率占 82 3 %。结论人工神经网络对药物的CZE的迁移时间可准确预测     

BP神经网络与遗传算法结合优化培养基配方应用研究

BP神经网络可以用来对培养基的配方进行优化设计,并且能够弥补一些传统方法的不足,但是由于BP神经网络自身固有的缺陷导致网络参数的不确定性,因而优化过程中常易陷于局部最优而想象全局最优结果的搜索.利用遗传算法全局寻优的特点,将其与BP神经网络结合,可以有效地提高网络的预测精度和推广能力.     

肝移植受者他克莫司血药浓度早期预测方案及评估

目的:建立人工神经网络用于估算他克莫司血药浓度。方法:收集36例肝移植受者口服他克莫司的150份稳态全血浓度数据,采用遗传算法配合动量法优化网络参数,建立人工神经网络。结果:人工神经网络平均预测误差(MPE)与平均绝对误差(MAE)分别为0.05±3.01ng·mL-1和2.09±2.12ng·mL-1,78.3%血药浓度数据绝对预测误差≤3.0ng·mL-1。人工神经网络准确性及精密度优于多元线性回归。结论:人工神经网络预测可用于预测他克莫司血药浓度。     

人工神经网络在类风湿关节炎诊断中的应用

目的:应用BP人工神经网络原理,设计一种类风湿关节炎疾病诊断的方法。方法:选用对类风湿关节炎敏感的8个指标,作为BP人工神经网络的输入数据,对样本进行训练和预测。结果:BP人工神经网络经通过对150例样本的运算,训练集的113例样本,训练正确率为97.4%;预测集的37例样本,预测正确率为91.9%。结论:BP人工神经网络能为类风湿关节炎作出较准确的诊断,能提高诊断的客观性。     

人工神经网络预测别嘌醇HP MC缓释片药物释放

目的用人工神经网络模型定量的预测HPMC的量和其固有黏度对药物释放的影响。方法以难溶性药物别嘌醇为模型药物,固定其他因素,HPMC的量和HPMC的固有黏度作为自变量,设计了18个处方并进行释放度检查;其中的13个处方作为训练处方,其他5个处方为验证处方,将上述的变量作为人工神经的输入,以药物在各个取样时间点的释放为输出,采用剔除一点交叉验证法建立人工神经网络模型。通过线性回归和相似因子说明人工神经网络的预测能力。结果训练和验证处方人工神经网络预测值与实际测定相符。结论建立BP人工神经网络,根据HPMC的量和其固有黏度可以定量的预测药物在各个时间点的药物释放。     

人工神经网络法预测肾移植术后患者环孢素A的血药浓度

目的：探讨人工神经网络技术用于肾移植术后患者环孢素A（cyclosporine A，CYA）血浓度预测的可行性。方法：收集60例肾移植术后患者CYA血药浓度监测结果及监测当日身高、体质量、肝肾功能等13项相关指标，采用EasyNN-plus8．0f软件用于人工神经网络的训练和模型的建立，计算方法为反向传播算（Back-Propagation）。用建立好的人工神经网络预测肾移植术后患者CYA血药浓度，同时采用多元线性回归法进行预测，并对两者的预测结果进行比较。结果：人工神经网络技术预测的血药浓度和观察值之间的相关系数为0．9314，多元线性回归的相关系数为0.7045，人工神经网络技术优于多元线性回归。结论：相对于传统的统计方法，人工神经网络技术在肾移植术后患者CYA血药浓度预测方面显示出良好的效能，但该方法也有待完善。     

基于神经网络的方法预测生物活性介质网络调控矽肺纤维化的研究

目的:探讨矽肺纤维化同生物活性介质之间的关系。方法:利用Delphi语言编制了BP人工神经网络模型计算机程序,建立并分析了矽肺胶原纤维预测的数学模型。结果:选定网络隐含层节点为9,初始权值阈值约为(-0.2,0.2),最大相对误差为4%,最小相对误差为0.2%。结论:应用神经网络具有较好的预测效果,可为临床医学研究提供一个很好的研究思路。     

Empirical versus mechanistic modelling: Comparison of an artificial neural network to a mechanistically based model for quantitative structure pharmacokinetic relationships of a homologous series of barbiturates

The aim of the current study was to compare the predictive performance of a mechanistically based model and an empirical artificial neural network (ANN) model to describe the relationship between the tissue-to-unbound plasma concentration ratios (Kpu's) of 14 rat tissues and the lipophilicity (LogP) of a series of nine 5-n-alkyl-5-ethyl barbituric acids. The mechanistic model comprised the water content, binding capacity, number of the binding sites, and binding association constant of each tissue. A backpropagation ANN with 2 hidden layers (33 neurons in the first layer, 9 neurons in the second) was used for the comparison. The network was trained by an algorithm with adaptive momentum and learning rate, programmed using the ANN Toolbox of MATLAB. The predictive performance of both models was evaluated using a leave-one-out procedure and computation of both the mean prediction error (ME, showing the prediction bias) and the mean squared prediction error (MSE, showing the prediction accuracy). The ME of the mechanistic model was 18% (range, 20 to 57%), indicating a tendency for overprediction; the MSE is 32% (range, 6 to 104%). The ANN had almost no bias: the ME was 2% (range, 36 to 64%) and had greater precision than the mechanistic model, MSE 18% (range, 4 to 70%). Generally, neither model appeared to be a significantly better predictor of the Kpu's in the rat.     

7种化合物细胞毒性与小鼠急性毒性的相关性研究

目的：为减少实验动物的使用,利用化学物质的体外细胞毒性数据对体内急性毒性进行预测。方法：MTT比色法检测7种新化学实体对CHL细胞的毒性作用,利用RC（Registry of Cytotox-icity）预测模型对急性毒性LD50值进行预测,并使用小鼠急性毒性上下法进行验证。结果：各化合物（1～7）细胞毒性IC50值分别为0.43、0.49、0.18、0.67、3.03、1.68、1.79mg/mL;根据RC预测模型,急性毒性LD50的预测值分别为2376.4、2478.3、1574.8、2087.6、4897.3、3331.8、3300.7mg/kg。经上下法检测,4号化合物的LD50值为1634.0mg/kg,其余6种化合物的LD50值均大于2000.0mg/kg。分别以预测值和实测值为依据对化合物毒性进行分级,二者相比,仅3号和4号化合物毒性分级略有差异,其它5种化合物的毒性分级基本一致。结论：体外细胞毒性数据可用来预测体内急性毒性,减少实验动物使用。     

Structure-toxicity relationship of aliphatic nitriles

Acute toxicity, in vitro metabolism and structure-toxicity relationship of aliphatic mononitriles were examined in mice pretreated with carbon tetrachloride (CCl4). The LD50 in mice pretreated with CCl4 (LD50-CCl4) was increased in most nitriles compared to that in untreated animals (LD50-cont.) with different degrees among compounds. Microsomal metabolism of nitriles to CN- was completely inhibited when microsomes were prepared from livers of mice pretreated with CCl4. Log (1/LD50-CCl4) was a linear function of partition coefficient, log P, i.e., log (1/LD50-CCl4) = -0.371 log P-0.152, and the equation was statistically significant (P less than 0.01), provided aceto- and 3-hydroxypropionitrile were omitted from the analysis. These two compounds seem to be different from the others in exerting the biological effect.     

Clinical application of artificial neural network (ANN) modeling to predict pharmacokinetic parameters of severely ill patients

Artificial neural network (ANN) modeling was used to evaluate the pharmacokinetics of aminoglycosides (arbekacin sulfate and amikacin sulfate) in severely ill patients. The plasma level was predicted by ANN modeling using parameters related to the severity of the patient's condition and the predictive performance was shown to be better than could be achieved using multiple regression analysis. These results indicate that there is a non-linear relationship between the pharmacokinetics of aminoglycosides and the severity of the patient's condition, and this should be taken into account when determining the dose for severely ill patients. Patients whose plasma levels are likely to fall below the effective level can be identified by ANN modeling with a predictive sensitivity and specificity superior to multivariate logistic regression analysis. The predictable range should be inferred from the data structure before the modeling in order to improve the predictive performance. The volume of distribution (Vd) in the normal range was weakly predicted by ANN modeling from the patients' data. Prediction of clearance by ANN modeling was poorer than that obtained from serum creatinine concentration by linear regression analysis. These results suggest that the input-output relationship (linear or non-linear) should be taken into account in selecting the modeling method. Linear modeling can give better predictive performance for linear systems and non-linear modeling can give better predictive performance for non-linear systems. In general, the performance of ANN modeling was superior to linear modeling for PK/PD prediction. For accurate modeling, a predictable range should be inferred from the data structure before the analysis. Restriction of the predictable region, as determined from the data structure, produced an increase in prediction performance. When applying ANN modeling in clinical settings, the predictive performance and predictable region should be investigated in detail to avoid the risk of harm to severely ill patients.     

Development of a QSAR for worst case estimates of acute toxicity of chemically reactive compounds

Future EU legislations enforce a fast hazard and risk assessment of thousands of existing chemicals. If conducted by means of present data requirements, this assessment will use a huge number of test animals and will be neither cost nor time effective. The purpose of the current research was to develop methods to increase the acceptability of in vitro data for classification and labelling regarding acute toxicity. For this purpose, a large existing database containing in vitro and in vivo data was analysed. For more than 300 compounds in the database, relations between in vitro cytotoxicity and rat or mouse intravenous and oral in vivo LD50 values were re-evaluated and the possibilities for definition of mechanism based chemical subclasses were investigated.

A high in vitro–in vivo correlation was found for chemicals classified as irritants. This can be explained by a shared unspecific cytotoxicity of these compounds which will act as the predominant mode of action for both endpoints, irritation and acute toxicity. For this subclass, which covered almost 40% of all compounds in the database, the LD50 values after intravenous dosing could be predicted with high accuracy. A somewhat lower accuracy was found for the prediction of oral LD50 values based on in vitro cytotoxicity data.

Based on this successful correlation, a classification and labelling scheme was developed, that includes a hazard based definition of the applicability domain (irritants) and a prediction of the labelling of compounds for their acute iv and oral toxicity. The scheme was tested by an external validation.     

Sirc-cvs cytotoxicity test: an alternative for predicting rodent acute systemic toxicity

An in vitro crystal violet staining method using the rabbit cornea-derived cell line (SIRC-CVS) has been developed as an alternative to predict acute systemic toxicity in rodents. Seventy-nine chemicals, the in vitro cytotoxicity of which was already reported by the Multicenter Evaluation of In vitro Toxicity (MEIC) and ICCVAM/ECVAM, were selected as test compounds. The cells were incubated with the chemicals for 72 hrs and the IC(50) and IC(35) values (microg/mL) were obtained. The results were compared to the in vivo (rat or mouse) "most toxic" oral, intraperitoneal, subcutaneous and intravenous LD(50) values (mg/kg) taken from the RTECS database for each of the chemicals by using Pearson's correlation statistics. The following parameters were calculated: accuracy, sensitivity, specificity, prevalence, positive predictability, and negative predictability. Good linear correlations (Pearson's coefficient; r>0.6) were observed between either the IC(50) or the IC(35) values and all the LD(50) values. Among them, a statistically significant high correlation (r=0.8102, p<0.001) required for acute systemic toxicity prediction was obtained between the IC(50) values and the oral LD(50) values. By using the cut-off concentrations of 2,000 mg/kg (LD(50)) and 4,225 microg/mL (IC(50)), no false negatives were observed, and the accuracy was 84.8%. From this, it is concluded that this method could be used to predict the acute systemic toxicity potential of chemicals in rodents.     

肝移植患者他克莫司个体化给药研究

目的:建立人工神经网络用于估算他克莫司血药浓度.方法:收集26例肝移植患者口服他克莫司的94份全血浓度数据,采用遗传算法配合动量法优化网络参数,建立人工神经网络.结果:人工神经网络平均预测误差(MPE)与平均绝对预测误差( MAE)分别为(-0.11±2.81) ng/mL和(2.14±1.72) ng/mL,78.6％血药浓度数据绝对预测误差≤3.0 ng/mL.多元线性回归 MPE与 MAE分别为(0.56±2.70) ng/mL和(2.15±1.63) ng/mL,9例次(9/14,64.3％)绝对预测误差≤3.0 ng/mL.人工神经网络准确性及精密度优于多元线性回归.结论:人工神经网络预测可用于预测他克莫司血药浓度,指导个体化给药.     

QSRR-based estimation of the retention time of opiate and sedative drugs by comprehensive two-dimensional gas chromatography

The quantitative structure-retention relationship (QSRR) of 69 opiate and sedative drugs against the comprehensive two-dimensional gas chromatography retention time (RT) was studied. The genetic algorithm (GA) was employed to select the variables that resulted in the best-fitted models. After the variables were selected, the linear multivariate regressions [e.g., the multiple linear regression (MLR), the partial least squares (PLS)] as well as the nonlinear regressions [e.g., the kernel PLS (KPLS), Levenberg–Marquardt artificial neural network (L–M ANN)] were utilized to construct the linear and nonlinear QSRR models. The correlation coefficient LGO-CV (Q²) of prediction for the GA-KPLS and L–M ANN models for training and test sets were (0.921 and 0.960) and (0.892 and 0.925), respectively, revealing the reliability of these models. The obtained results using L-M ANN were compared with those of GA-MLR, GA-PLS, and GA-KPLS, exhibiting that the L–M ANN model demonstrated a better performance than that of the other models. The resulting data indicated that L–M ANN could be used as a powerful modeling tool for the QSRR studies. This is the first research on the QSRR of the drug compounds against the RT using the GA-KPLS and L–M ANN.