PLS-DA分析R实现

偏最小二乘法(Partial Least Squares)最初不是为分类和辨别问题而设计的，但经常被用于这一目的。响应矩阵Y是定性的，在内部被重新编码为一个虚拟块矩阵，记录每个观察变量的成员资格，即每个响应类别都通过一个指标变量进行编码。然后，PLS回归（现在的PLS-DA）就像Y是一个连续矩阵一样运行，并且在线性判别分析面临碰撞问题的大型数据集上运行良好。

PLS-判别分析（PLS-DA）是一个线性分类模型，能够预测新样本的类别。
稀疏PLS-DA（sPLS-DA）能够选择数据中最具预测性或判别性的特征，帮助对样本进行分类。

方法一：mixOmics包中的plsda函数 

install.packages("mixOmics")

library(mixOmics)

plsda函数，该函数无法得到R2及Q2。

data(breast.tumors)
X <- breast.tumors$gene.exp
Y <- breast.tumors$sample$treatment

plsda.breast <- plsda(X, Y, ncomp = 2)
plotIndiv(plsda.breast, ind.names = TRUE, ellipse = TRUE, legend = TRUE)

得到VIP值

library(RVAideMemoire)
PLSDA.VIP(plsda.breast, graph = TRUE)

 

方法二： DiscriMiner包中的plsDA函数

执行部分最小二乘法（PLS）判别分析，可选择包括随机离开-k折叠交叉验证。

## Not run: 
# load iris dataset
data(iris)

# PLS discriminant analysis specifying number of components = 2
my_pls1 = plsDA(iris[,1:4], iris$Species, autosel=FALSE, comps=2)
my_pls1$confusion
my_pls1$error_rate
# plot circle of correlations
plot(my_pls1)

# PLS discriminant analysis with automatic selection of components
my_pls2 = plsDA(iris[,1:4], iris$Species, autosel=TRUE)
my_pls2$confusion
my_pls2$error_rate

# linear discriminant analysis with learn-test validation
learning = c(1:40, 51:90, 101:140)
testing = c(41:50, 91:100, 141:150)
my_pls3 = plsDA(iris[,1:4], iris$Species, validation="learntest",
learn=learning, test=testing)
my_pls3$confusion
my_pls3$error_rate

my_pls3$R2

         R2X    R2Xcum        R2Y    R2Ycum
t1 0.7332792 0.7332792 0.47105647 0.4710565
t2 0.2223789 0.9556582 0.07414135 0.5451978

my_pls3$Q2

Q2.setosa Q2.versicolor Q2.virginica   Q2.global
t1 0.86919713   0.046061518   0.48123542  0.46549802
t2 0.18666695   0.143015160   0.08398902  0.12743855
t3 0.06393944   0.003635849   0.09319238  0.03866148
t4 0.01852317  -0.015969033  -0.02068680 -0.01891103

my_pls3$VIP

             Component 1 Component 2 Model VIP
Sepal.Length   0.9271868   0.8856696 0.8856696
Sepal.Width    0.6638666   0.9222639 0.9222639
Petal.Length   1.1682002   1.0874663 1.0874663
Petal.Width    1.1553848   1.0873986 1.0873986

plot(my_pls3)

 

ref:

plsDA: PLS Discriminant Analysis in DiscriMiner: Tools of the Trade for Discriminant Analysis (rdrr.io)