1.背景知识
这是一篇2018年的22分文章，《The Immune Landscape of Cancer》将肿瘤样本划分为六种免疫亚型:wound healing, IFN-γ dominant, inflammatory, lymphocyte depleted, immunologically quiet, and TGF-β dominant,R包ImmuneSubtypeClassifier可以根据基因表达矩阵来对样本划分亚型。
多说一句，TCGA 样本的免疫亚型不需要自己计算，在文章附表2中已经有了。那就用于其他来源的表达矩阵咯。
2.示例数据和代码
出自R包的readme文件
#devtools::install_github("Gibbsdavidl/ImmuneSubtypeClassifier")
library(readr)
library(ImmuneSubtypeClassifier)
download.file(url = 'https://raw.githubusercontent.com/CRI-iAtlas/shiny-iatlas/develop/data/ebpp_test1_1to20.tsv', destfile = 'ebpp_test.tsv')
dat <- read_tsv('ebpp_test.tsv')
dat2 <- as.data.frame(dat[!duplicated(dat$GeneID),])
Xmat <- dat2[,-1]
rownames(Xmat) <- dat2[,1]
Xmat[101:104,1:4]
## XY1 XY2 XY3 XY4
## ABCE1 1375.600 1202.25 953.933 909.373
## ABCF1 1239.670 1664.88 1192.710 635.941
## ABCF2 978.866 1590.79 2497.410 1848.180
## ABCF3 8834.770 3333.38 2428.680 5329.980
Xmat就是作者最终整理成的输入数据了。是一个未经log的表达矩阵
res0 <- callEnsemble(X = Xmat, geneids = 'symbol')
res0
## SampleIDs BestCall 1 2 3 4
## 1 XY1 4 1.121173e-07 1.873900e-06 6.311234e-02 0.9027497470
## 2 XY2 3 4.243225e-02 2.309184e-06 5.495564e-01 0.0084770960
## 3 XY3 4 3.095071e-04 1.029907e-06 1.635270e-01 0.9063920975
## 4 XY4 4 1.154656e-04 3.823049e-07 2.888787e-02 0.8831390142
## 5 XY5 4 1.193054e-07 8.741189e-06 5.060996e-01 0.9260828793
## 6 XY6 4 2.479082e-04 5.528710e-05 1.183165e-04 0.9923758209
## 7 XY7 6 9.421768e-03 1.079501e-03 1.000559e-01 0.0084094275
## 8 XY8 3 6.143126e-04 2.594000e-06 8.872822e-01 0.1306741871
## 9 XY9 4 1.001003e-04 4.479314e-06 3.989228e-03 0.9793768525
## 10 XY10 2 4.624279e-06 9.888145e-01 7.359737e-06 0.0053910189
## 11 XY11 4 5.837736e-05 7.877929e-06 1.970483e-03 0.9895575941
## 12 XY12 3 1.944198e-06 2.005060e-06 3.616153e-01 0.5258071870
## 13 XY13 4 9.002434e-07 9.563100e-04 2.673559e-06 0.9931332767
## 14 XY14 4 1.235332e-05 3.907399e-05 2.785671e-03 0.9972651005
## 15 XY15 4 4.831095e-05 1.127145e-04 1.685999e-04 0.9935480356
## 16 XY16 3 2.731656e-03 1.722274e-05 9.742068e-01 0.0007705171
## 17 XY17 4 3.431090e-06 1.327718e-03 1.570120e-05 0.9971910715
## 18 XY18 3 3.816116e-07 5.000733e-03 8.685111e-01 0.0030849737
## 19 XY19 4 4.787456e-05 2.856209e-06 5.303099e-01 0.9858489633
## 20 XY20 4 9.208488e-04 1.361713e-04 3.989896e-04 0.9504880905
## 5 6
## 1 5.132385e-02 5.121503e-05
## 2 1.000665e-04 2.261875e-04
## 3 5.731729e-03 1.741630e-04
## 4 1.236814e-03 7.847964e-05
## 5 1.232723e-02 5.358944e-04
## 6 1.019272e-03 2.854635e-04
## 7 3.649302e-05 6.113418e-01
## 8 8.199657e-04 1.142911e-04
## 9 2.182218e-03 1.554600e-04
## 10 3.058250e-05 6.004089e-04
## 11 4.063185e-03 6.252101e-04
## 12 1.320550e-02 2.124778e-04
## 13 1.996231e-04 1.273039e-04
## 14 2.466791e-03 5.618507e-04
## 15 4.465038e-03 2.503012e-04
## 16 4.555504e-05 6.263918e-03
## 17 3.616062e-05 1.134532e-04
## 18 2.343850e-04 2.590704e-03
## 19 2.194258e-02 9.931145e-05
## 20 1.666186e-03 1.220569e-02
3.对输入数据的探索
这个包的资料特别少，也没有对应的文章。翻了帮助文档和github，各种教程，没有找到对输入数据的要求说明，暂不知道用的是TPM，FPKM，cpm还是别的什么类型。只能看出没有取log。如果后续有进展，我会在语雀中更新，点击文末原文链接就能看到。
我想试试看log后的表达矩阵是否也可以作为输入
res1 <- callEnsemble(X = log2(Xmat+1), geneids = 'symbol')
res1
## SampleIDs BestCall 1 2 3 4
## 1 XY1 4 1.121173e-07 1.873900e-06 6.311234e-02 0.9027497470
## 2 XY2 3 4.243225e-02 2.309184e-06 5.495564e-01 0.0084770960
## 3 XY3 4 3.095071e-04 1.029907e-06 1.635270e-01 0.9063920975
## 4 XY4 4 1.154656e-04 3.823049e-07 2.888787e-02 0.8831390142
## 5 XY5 4 1.193054e-07 8.741189e-06 5.060996e-01 0.9260828793
## 6 XY6 4 2.479082e-04 5.528710e-05 1.183165e-04 0.9923758209
## 7 XY7 6 9.421768e-03 1.079501e-03 1.000559e-01 0.0084094275
## 8 XY8 3 6.143126e-04 2.594000e-06 8.872822e-01 0.1306741871
## 9 XY9 4 1.001003e-04 4.479314e-06 3.989228e-03 0.9793768525
## 10 XY10 2 4.624279e-06 9.888145e-01 7.359737e-06 0.0053910189
## 11 XY11 4 5.837736e-05 7.877929e-06 1.970483e-03 0.9895575941
## 12 XY12 3 1.944198e-06 2.005060e-06 3.616153e-01 0.5258071870
## 13 XY13 4 9.002434e-07 9.563100e-04 2.673559e-06 0.9931332767
## 14 XY14 4 1.235332e-05 3.907399e-05 2.785671e-03 0.9972651005
## 15 XY15 4 4.831095e-05 1.127145e-04 1.685999e-04 0.9935480356
## 16 XY16 3 2.731656e-03 1.722274e-05 9.742068e-01 0.0007705171
## 17 XY17 4 3.431090e-06 1.327718e-03 1.570120e-05 0.9971910715
## 18 XY18 3 3.816116e-07 5.000733e-03 8.685111e-01 0.0030849737
## 19 XY19 4 4.787456e-05 2.856209e-06 5.303099e-01 0.9858489633
## 20 XY20 4 9.208488e-04 1.361713e-04 3.989896e-04 0.9504880905
## 5 6
## 1 5.132385e-02 5.121503e-05
## 2 1.000665e-04 2.261875e-04
## 3 5.731729e-03 1.741630e-04
## 4 1.236814e-03 7.847964e-05
## 5 1.232723e-02 5.358944e-04
## 6 1.019272e-03 2.854635e-04
## 7 3.649302e-05 6.113418e-01
## 8 8.199657e-04 1.142911e-04
## 9 2.182218e-03 1.554600e-04
## 10 3.058250e-05 6.004089e-04
## 11 4.063185e-03 6.252101e-04
## 12 1.320550e-02 2.124778e-04
## 13 1.996231e-04 1.273039e-04
## 14 2.466791e-03 5.618507e-04
## 15 4.465038e-03 2.503012e-04
## 16 4.555504e-05 6.263918e-03
## 17 3.616062e-05 1.134532e-04
## 18 2.343850e-04 2.590704e-03
## 19 2.194258e-02 9.931145e-05
## 20 1.666186e-03 1.220569e-02
identical(res0$BestCall,res1$BestCall)
## [1] TRUE
两次计算出来的结果一样，那就是可以咯
再试试标准化后的数据行不行
res2 <- callEnsemble(X = t(scale(t(log2(Xmat+1)))), geneids = 'symbol')
res2
## SampleIDs BestCall 1 2 3 4
## 1 XY1 2 1.055616e-04 9.624696e-01 1.735410e-01 4.088800e-01
## 2 XY2 3 5.200884e-02 2.582083e-01 2.077989e-01 6.790864e-04
## 3 XY3 3 1.050717e-01 8.588433e-05 1.228831e-01 1.287199e-02
## 4 XY4 2 2.210707e-03 1.608360e-01 3.885327e-04 3.327382e-02
## 5 XY5 3 3.972800e-06 9.615958e-03 6.854322e-01 2.701341e-01
## 6 XY6 4 1.715342e-04 3.799206e-01 4.371099e-06 8.692024e-01
## 7 XY7 6 1.250667e-01 1.097162e-01 4.140158e-03 6.115460e-04
## 8 XY8 3 2.792812e-04 2.527673e-04 8.357488e-01 1.747175e-02
## 9 XY9 4 9.675489e-01 8.905539e-02 1.780803e-04 9.530877e-01
## 10 XY10 2 1.756920e-07 9.999766e-01 5.824814e-05 1.829826e-03
## 11 XY11 4 2.118116e-07 1.764702e-02 1.047677e-04 6.112984e-01
## 12 XY12 3 2.059111e-07 1.993342e-01 6.549236e-01 1.061971e-02
## 13 XY13 2 4.458124e-07 9.985121e-01 7.005165e-08 1.838380e-02
## 14 XY14 4 9.115437e-06 2.641302e-03 6.308785e-06 9.922896e-01
## 15 XY15 2 7.724871e-07 9.421750e-01 9.746566e-07 9.982958e-01
## 16 XY16 3 3.725465e-06 2.441265e-03 5.236077e-01 3.568041e-06
## 17 XY17 2 8.232909e-06 9.958895e-01 4.845522e-07 4.058149e-02
## 18 XY18 3 5.211974e-07 3.527417e-01 9.039713e-01 3.747310e-03
## 19 XY19 2 9.518531e-06 1.477052e-01 2.665034e-03 1.926993e-01
## 20 XY20 4 1.538433e-01 1.374537e-01 2.018919e-06 3.616710e-01
## 5 6
## 1 0.0970802158 0.0054663994
## 2 0.0022650487 0.0541812675
## 3 0.0191998631 0.0012868469
## 4 0.0070779554 0.0004822279
## 5 0.0083340211 0.0135210226
## 6 0.0007782122 0.0006131011
## 7 0.0004781231 0.1474968009
## 8 0.0134209383 0.0001762331
## 9 0.0017751575 0.0001619612
## 10 0.0004091656 0.0014408990
## 11 0.0018340806 0.0065515421
## 12 0.0795782432 0.0040155074
## 13 0.0009391853 0.0014942663
## 14 0.0016644751 0.0035965345
## 15 0.0084981914 0.0013056903
## 16 0.0001257564 0.1152272336
## 17 0.0021467893 0.0009310142
## 18 0.0043335482 0.0023881147
## 19 0.0033517949 0.0070664277
## 20 0.0082207613 0.0109260329
identical(res0$BestCall,res2$BestCall)
## [1] FALSE
head(res0$BestCall)
## [1] 4 3 4 4 4 4
head(res2$BestCall)
## [1] 2 3 3 2 3 4
结果不一样了，那就是不行
4.errorreport
区分免疫亚型用到了485个关键的基因。geneMatchErrorReport告诉你有多少基因是在输入数据中不存在的，如果比例过高，那最终计算出来的亚型就不那么可靠。
data(ebpp_gene)
head(ebpp_genes_sig)
## Symbol Entrez Ensembl
## 235 ACTL6A 86 ENSG00000136518
## 294 ADAM9 8754 ENSG00000168615
## 305 ADAMTS1 9510 ENSG00000154734
## 322 ADAR 103 ENSG00000160710
## 340 ADCY7 113 ENSG00000121281
## 479 AIMP2 7965 ENSG00000106305
geneMatchErrorReport(X=Xmat, geneid='symbol')
## $matchError
## [1] 0
##
## $missingGenes
## [1] Symbol Entrez Ensembl
## <0 rows> (or 0-length row.names)
5.可视化
原数据是没有分组的，为了体现不同组的亚型比较，这里加上一列分组，仅做例子。
dat = res0
dat$BestCall = factor(dat$BestCall)
dat$group = rep(c("a","b"),each = 10)
library(ggplot2)
library(paletteer)
ggplot(data = dat,aes(x = group))+
geom_bar(aes(fill = BestCall),
position = position_dodge2(padding = 0,
preserve = "single"))+
theme_classic()+
scale_fill_paletteer_d("RColorBrewer::Set2")
library(dplyr)
dat2 <- dat %>%
group_by(group, BestCall) %>%
summarise(count = n())
ggplot() + geom_bar(data =dat2, aes(x = group, y = count, fill = BestCall),
stat = "identity",
position = "fill")+
theme_classic()+
scale_fill_paletteer_d("RColorBrewer::Set2")