Prediction of biomarkers and therapeutic combinations for anti-PD-1 immunotherapy using the global gene network association
Owing to a lack of response to the anti-PD1 therapy for most cancer patients, we developed a network approach to infer genes, pathways, and potential therapeutic combinations that are associated with tumor response to anti-PD1. First, the network guilt-by-association method was used to calculate functional relatedness (hereafter termed MHC I association score) of each genes with the MHC I pathway genes (Fig. 1A). Pathways that enrich genes with high association scores would be associated with the anti-PD1 (Fig. 1A).Second, by leveraging gene expression data of samples, we were able to identify genes and pathways that are associated with the anti-PD1 in a specific cancer type or a specific cellular condition (Fig. 1B). Third, TCGA transcriptomic data can be integrated with the top MHC I-associated genes to select the gene signature and calculate the MHC I association immunoscore (MIAS) to predict patient response to anti-PD1 for a given cancer type. Forth, mining TCGA genetic data using the top MHC I-associated genes can also explore genetic and epigenetic aberrations associated with anti-PD-1 in a given cancer type. Fifth, mapping the drug target data to the top MHC I-associated genes can identify compounds that could enhance tumor response to anti-PD1.
Fig. 1: The network-based approach
Our prediction successfully identified genes and pathways known to be associated with anti-PD1, and was further validated by 6 CRISPR gene sets associated with tumor resistance to cytotoxic T cells and targets of the 36 compounds that have been tested in clinical trials for combination treatments with anti-PD1. Mapping drug target data to our top MHC I-associated genes also identified inhibitors that could potentially enhance tumor response to anti-PD1, such as inhibitors of CDK, GSK3B, and PTK2. The MIAS score showed a good correlation with patient response to anti-PD1 for 411 melanoma samples complied from 6 cohorts.
Our analysis indicated that MIAS score can be a useful feature to build integrative machine-learning models for anti-PD1 response prediction. The following script help people to calculate the MIAS scores of the melanoma patient samples using their transcriptomic data.
-
Download the file "MIAS_SKCM.zip" from this site. After unzipping the "MIAS_SKCM.zip", navigate to the extension folder, "MIAS_SKCM", and set it as the working directory
-
Load the require R packages and scripts
library(GSVA)
source("MHC_functions.r")
- Load the selected SKCM signature genes for calculating MIAS scores (The signature genes were selected by integrative analysis of our MHC I-association prediction and TCGA SKCM transcriptomic data).
filein="./data/Table.S6_Immune_positive signature.xls"
DataM1 <- read.table(filein,sep="\t",header=T, quote="")
Signatures_M<-NULL
Signatures_M[[1]]=as.character(DataM1[[1]])
- load the gene expressio data of the melanoma patient samples for response prediction. The gene expression value should be calucated as transcripts per million (TPM) that normalize counts for library size and gene length. Here, an example dataset (complied from Riaz, et al 2017) was loaded. This data set contains a gene expression matrix od patient samples (DataM_EX), patient response (Response), biopsy timepoint (PreOn).
load("./data/Example.Data_Riaz.2017.RData")
- the MIAS scores of samples can be calculated using the signature (Signatures_M) and the gene expresison matrix (DataM_EX )
MIAS_Score.Pre<-MIAS.Score.GSVA(DataM_EX,Signatures_M)
- Quantify the correlation between the MIAS score and patient response using the AUC value of the ROC curve and Wilcox test (only appicable to samples with response data)
outcome=rep(1,length(MIAS_Score))
outcome[which(Response=="NR")]=0
ROC<-ROCF(outcome,MIAS_Score)
score.pos<-MIAS_Score[which(Response=="R")]
score.neg<-MIAS_Score[which(Response=="NR")]
Wilcox.pval<-as.numeric(wilcox.test(score.pos, score.neg, alternative="greater")$p.value)
print(ROC$AUC)
print(Wilcox.pval)
You can use the calculated MIAS score to build up a model to predict patient response to anti-PD1 in melanoma.
Our analysis also showed that the integration of the MIAS and IMPRES scores also have a better prediction performance than the two individual method (Fig. 6D in the manuscript). We thus used the MIAS and IMPRES scores of the collected 411 melanoma samples as the data fetures to train two predictors of anti-PD1 response using support vector machine (SVM) respectively for pre- and on-treated melanoma patient samples. These two predictors can help people to predict responses of SKCM patient samples directly using their transcriptomic data.
-
Download the file "MIAS_SKCM.zip" from this site. After unzipping the "MIAS_SKCM.zip", navigate to the extension folder, "MIAS_SKCM", and set it as the working directory
-
Load the require R packages
library(e1071) #svm
library(GSVA)
source("MHC_functions.r")
- Load the data file containing the 2 pre-trained predictors: MIAS.IMPRES.Classifier_Pre, and MIAS.IMPRES.Classifier_On.
load("./data/Response.Predictors_PD1_SKCM.RData")
- Load the gene expressio data of the melanoma patient samples for response prediction. Here, an example dataset (Riaz, et al 2017) was loaded.
load("./data/Example.Data_Riaz.2017.RData")
- Seperate data of pre- and on-treatment samples
DataM_EX.Pre<-DataM_EX[,which(PreOn=="Pre")]
Response.Pre<-Response[which(PreOn=="Pre")]
DataM_EX.On<-DataM_EX[,which(PreOn=="On")]
Response.On<-Response[which(PreOn=="On")]
- Load the selected SKCM signature genes for calculating MIAS scores.
filein="./data/Table.S6_Immune_positive signature.xls"
DataM1 <- read.table(filein,sep="\t",header=T, quote="")
Signatures_M<-NULL
Signatures_M[[1]]=as.character(DataM1[[1]])
- Calculate the MIAS and IMPRES Scores of the pre- and on-treatment samples
MIAS_Score.Pre<-MIAS.Score.GSVA(DataM_EX.Pre,Signatures_M)
IMPRES_Score.Pre<-IMPRES.Score(DataM_EX.Pre)
MIAS_Score.On<-MIAS.Score.GSVA(DataM_EX.On,Signatures_M)
IMPRES_Score.On<-IMPRES.Score(DataM_EX.On)
- Response prediction of the on-treatment samples using the on-treatment predictor that were trained using both of MIAS and IMPRES scores (MIAS.IMPRES.Classifier_On).
data_set.On=data.frame(MIAS=MIAS_Score.On,IMPRES=IMPRES_Score.On,Response=Response.On)
prediction.On = predict(MIAS.IMPRES.Classifier_On, newdata = data_set.On[-3], probability =T)
- Print out the predicted patient responses
print (prediction.On)
We can also do the response prediction to the pre-treatment samples using the pre-treatment predictor, MIAS.IMPRES.Classifier_Pre (the script was not listed here).