self-organizing map (SOM) 是一種被無(wú)監(jiān)督學(xué)習(xí)訓(xùn)練產(chǎn)生的一個(gè)低維（通常為二維)的人工神經(jīng)網(wǎng)絡(luò)，可作為一種進(jìn)行降維的方法。在高通量測(cè)序數(shù)據(jù)中可作為一種將不同樣本的特征聚類的可視化方式。
R 實(shí)現(xiàn)：oposSOM包
oposSOM軟件包僅需要輸入以基因表達(dá)矩陣數(shù)據(jù)，例如經(jīng)過(guò)標(biāo)準(zhǔn)化處理microarray 數(shù)據(jù)或RNA-seq數(shù)據(jù)。
處理過(guò)程：
? The SOM space obtained from the training process is characterized by several supporting maps and profiles providing, e.g. the number of genes mapped to each meta-gene.
? Samples are individually portrayed in PDF report sheets allowing
the detailed examination of their expression landscapes and especially to identify modules of co-expressed genes.
? Feature maps, reports and lists allow feature selection and evaluation of their statistical significance.
? Gene set enrichment analysis of the expression modules provides
their functional context based on a large collection of predefined
gene sets.
? Sample diversity analysis and class discovery is performed using
multiple algorithms (e.g. hierarchical clustering, correlation
spanning tree) and different metrics (Euclidean distance,
Pearson’s correlation coefficient).
1.安裝oposSOM 包
if (!requireNamespace("BiocManager", quietly = TRUE))
install.packages("BiocManager")
BiocManager::install("oposSOM")
library(oposSOM)
env <- opossom.new(list(dataset.name="Tissues",dim.1stLvlSom=20))
data(opossom.tissues)
env$indata <- opossom.tissues
env$group.labels <- c(rep("Homeostasis", 2),"Endocrine","Digestion","Exocrine","Epithelium","Reproduction","Muscle",rep("Immune System", 2),rep("Nervous System", 2) )
env$group.colors <- c(rep("gold", 2),"red2","brown","purple","cyan","pink","green2",rep("blue2", 2),rep("gray", 2) )
opossom.run(env)
具體介紹可見(jiàn)：The oposSOM Package
參考：
https://zhuanlan.zhihu.com/p/73534694
https://www.fmi.uni-leipzig.de/Media/DissAbstracts/abstract.wirth.pdf
