核磁共振波谱-13讲述.ppt

整体样本总的代谢趋势和轮廓 排除样本异常点 用于质量控制分析 个体差异 代谢轨迹 疾病监控 主成分分析的作用 偏最小二乘法-判别分析 Partial least squares-discriminant analysis, PLS-DA 贝叶斯概率论方法 Bayesian probabilistic approaches 线性差别法 Linear discriminant analysis, LDA 神经网络 Neural networks, NN 簇类独立软模式 Soft independent modeling of class analogy, SIMCA 偏最小二乘法 Partial least squares, PLS 有监督的模式识别 Supervised pattern recognition PLS-DA Supervised learning method. Principles that of PCA. But in PLS, a second piece of information is used, namely, the labeled set of class identities. Two data tables considered namely X input data from samples and Y containing qualitative values, such as class belonging, treatment of samples ? The quantitive relationship between the two tables is sought. X TPT + E Y TCT + E The PLS algorithm maximizes the covariance between the X variables and the Y variables PLS models negatively affected by systematic variation in the X matrix not related to the Y matrix not part of the joint correlation structure between X-Y. 偏最小二乘法-判别分析 Partial least squares-discriminant analysis, PLS-DA PLS-DA得分图及以UV换算的载荷图 OPLS-DA OPLS method is a recent modification of the PLS method to help overcome pitfalls Main idea to seperate systematic variation in X into two parts, one linearly related to Y and one unrelated orthogonal . Comprises two modeled variations, the Y-predictive TpPpT and the Y-orthogonal ToPoT compononents. Only Y-predictive variation used for modeling of Y. X TpPpT + ToPoT + E Y TpCpT + F E and F are the residual matrices of X and Y OPLS-DA compared to PLS-DA 正交偏最小二乘法-判别分析 orthogonal Partial least squares-discriminant analysis, OPLS-DA OPLS-DA得分图及以UV换算并经过回溯转换后的载荷图 control dosed Between group variation Within group variation OPLS-DA 组间的区分 寻找潜在的生物标志物 比较分析 对未知样本进行预测 O PLS-DA的作用 OPLS-DA中选择生物标志物的方式—VIP值法 Treated Control VIP临界值的确定 取所有VIP值最高的1%，10%或20% 取VIP值大于1 OPLS-DA中选择生物标志物的方式—相关系数法 control Treated 相关系数临界值的确定 相关系数临界值表（单变量统计分析） 由自己的样本数及分析结果确定合适的临界值 一般情况下，相关系数R 0.9为高度相关，0.75 R 9为显著相关，0.