PCA tutorial

2.PCA and feature selection.pptx

2.PCA and feature selection.r

# PCA start
  # I recommend that you set scale TRUE when doing PCA. Since variance influence the resulting PCs.
  # Center=TRUE is to make median=0 and scale.=TRUE change unit variance equal

  nc2=t(nc2) # Transpose the log2-CPM matrix, samples should be row and variables are on column

# Row : Samples

# Column : Variables such as genes

  pc=prcomp(nc2,center = T,scale. = T)
  print(pc) # You will see contribution values of each genes for individual Principal component
  pc1=sort(abs(pc$rotation[,1]),decreasing = T) # Sort the genes according their contribution value
  print(names(pc1[1:100])) # Top 100 contributing genes

  # 3d pca plot
  install.packages('rgl');library(rgl)
  pc$x
  cols=c(rep('red',3),rep('blue',3),rep('darkgreen',3),rep('orange',3),rep('purple',3))
  plot3d(x=pc$x[,1],y=pc$x[,2],z=pc$x[,3],col=cols,size=10,
       xlab='PC1',ylab='PC2',zlab='PC3') # x=PC1, y=PC2, z=PC3
  text3d(x=pc$x[,1],y=pc$x[,2],z=pc$x[,3], # Add labels for points
       text=c(rep('A',3),rep('B',3),rep('C',3),rep('D',3),rep('E',3)),add=T,size=10)
  rgl.postscript('your_PCA_plot.pdf','pdf') # Export what you see as PDF format
 
  # Rotate and see whether samples are separated well
  install.packages('pheatmap');library(pheatmap)
  pheatmap(nc[rownames(nc) %in% names(pc1[1:100]),],scale = 'row') # heatmap for PC1 top 100 genes
  # PCA end

# 새 데이터 들어왔을 때 기존 pca로 데이터 모델에 적용하는 법

predict(pc, newdata='새데이터 넣을것')

http://planspace.org/2013/02/03/pca-3d-visualization-and-clustering-in-r/