ikarus cancer cell classification in R

Preliminaries

Setup the conda environment

Here we install the appopriate python version, along with all of the required R and python packages.

conda create  -y -n ikar3.8 -python=3.8
conda install -y libgcc
conda install -y r-essentials r-base
conda install -y -c bioconda bioconductor-biocinstaller bioconductor-zellkonverter bioconductor-singlecellexperiment
conda install -y r-seurat
pip install git+https://github.com/BIMSBbioinfo/ikarus.git

Load the conda environment in R

Start the R interpreter, load reticulate, and load the conda environment.

conda_path should contain the path to the installed conda environment.

library(reticulate)
conda_path = c("~/bin/Software/miniconda3/envs/ikar3.8")
use_condaenv(conda_path)

Data preparation

Download the ikarus model

trained_model_path = "https://github.com/BIMSBbioinfo/ikarus/raw/master/tutorials/out/core_model.joblib"

download.file(trained_model_path, "core_model.joblib")

Download gene signatures

signatures_path = "https://github.com/BIMSBbioinfo/ikarus/raw/master/tutorials/out/signatures.gmt"
download.file(signatures_path, "signatures.gmt")

Run Ikarus

Download the adata

adata_path = "https://bimsbstatic.mdc-berlin.de/akalin/Ikarus/part_1/data/tirosh17_headneck/adata.h5ad"
download.file(adata_path, "tirosh_adata.h5ad")

Read the h5ad into an anndata object

anndata = import("anndata")
adata = anndata$read_h5ad("tirosh_adata.h5ad")

Load the trained ikarus model

ikarus = import("ikarus")
model  = ikarus$classifier$Ikarus(
     
  signatures_gmt = file.path("signatures.gmt"), 
  out_dir="ikarus_path"
)
model$load_core_model("core_model.joblib")

Predict the cell class

class_predict = model$predict(adata, "test")

Look at the prediction results

results = data.frame(
  true = adata$obs$tier_0,
  pred = class_predict
)
with(results, table(true, pred))

        pred
true     Normal Tumor
  Normal   3362     1
  Tumor       8  2207

MISC helpers

Seurat to SingleCellExperiment

sce = SingleCellExperiment(
  assays = list(counts = GetAssayData(seu, "counts", "RNA"),
  colData = DataFrame(seu@meta.data)
) 

Convert SingleCellExperiment to AnnData

library(basilisk)
adata = basiliskRun(fun = function(sce) {
     # Convert SCE to AnnData:
     SCE2AnnData(sce)
     
}, env = conda_path, sce = sce)

Process count data using ikarus

If the input adata object contains counts, the matrix can be processed using ikarus preprocess_adata function.

If the data is already normalized, no preprocessing is required

IMPORTANT the data should not be scaled!

adata = ikarus$data$preprocess_adata(adata)