This repository contains data and source code used to train the ILC2 XGBoost classifier in the manuscript PD-1 blockade unleashes ILC2-dependent anti-tumor immunity in melanoma by Jacquelot et al. The XGBoost model is provided as a RData object for easy application. Install xgboost package from CRAN in R >= 3.6.3, then:

Load pre-trained XGBoost model

library(xgboost)
load("tonsils_ILC2_xgb_model.RData")

Access classifier feature (gene) names by

head(bst$feature_names)
nFeatures <- length(bst$feature_names)

Apply the model to new datasets

To use the model to obtain ILC2 prediction scores on new datasets, apply TMM-normalisation to your RNAseq count data (optional), compute log2-cpms and subset your dataset on genes in the classifier (Note feature names are Homo-sapiens gene symbols). Here testdata is a matrix of logCPM values with nFeatures columns and nSamples rows

preds <- predict(bst, testdata)