This repository is designed to allow interested people to reproduce the results and figures of our paper called 'Observation weights unlock bulk RNA-seq tools for zero inflation and single-cell applications', currently on bioRxiv at https://www.biorxiv.org/content/early/2018/01/18/250126. All code in the repository is distributed under the GPL-3 license.
For examples on how to use the method, please see the zinbwave vignette or the zingeR vignette on how to use the observation weights in your analysis.
For any questions or issues with the code on this repository, please use the "Issues" tab.
To be able to run the code in this repo, it is required to have R (>=3.4) and the following packages.
- zingeR
- countsimQC
- Seurat
- ggplot2
- RColorBrewer
- devtools
- scales
- gridBase
- grid
- Matrix
- dplyr
- plyr
- cowplot
- Rtsne
- FNN
- rARPACK
- zinbwave
- limma
- edgeR
- DESeq2
- BiocParallel
- doParallel
- Biobase
- scde
- SingleCellExperiment
- MAST
- xCell
- fgsea
- clusterExperiment
- GSEABase
The functions used to estimate parameters based on real scRNA-seq data, and to simulate the expression counts can be found in the simulationHelpFunctions_v7_diffInZero.R file in the zingeRsimulationFunctions folder. This framework has been used to simulate all scRNA-seq datasets. We have also simulated a bulk RNA-seq dataset and code for this simulation can be found in the rnaseqSim.R file. The quality of the simulated datasets has been evaluated using the countsimQC package, and code for this evaluation can be found in the zinbwaveSimulations/evaluateSimulatedData folder. The code for the evaluations on the simulated Islam, Trapnell and 10X datasets can be found in the respective islam_sims_fc2, trapnell_sims_fc2 and tenX_sims_fc2 folders. In the respective files, the FDP-TPR plots are saved and the final Figures can be recreated with the fdrTprPlots.R file. We have investigated the effect of the penalty parameter on the simulated Islam and 10x datasets, which can respectively be found in the islam_sims_fc2_epsilon and tenX_sims_fc2_epsilon folders.
To generate the plots related to the false positive rate control, run FPR_mocks_tenx.Rmd for the 10X genomics dataset and FPR_mocks.Rmd for the Usoskin dataset. Additionally, to generate the plots related to PCER when the penalization parameter of ZINB-WaVE is varied, run FPR_mocks_eps_tenx.Rmd for the 10X genomics dataset and FPR_mocks_eps_usoskin.Rmd for the Usoskin dataset.
To generate the plots related to the analysis of the 10x Genomics PBMC dataset, first run createDataObject.Rmd to create the data files. Then, to generate the data when the clustering is done using PCA, run both de_seurat.Rmd and de_othermethods.Rmd. There are two files to run instead of one unique file because packages Seurat and zinbwave load both many packages and R complains that there are too many packages loaded. To generate the plots when the clustering is done using ZINB-WaVE, run dimredZinbwave.Rmd, clusterW.Rmd, and then de.Rmd. Finally, to generate the plots, run plotPaper.Rmd.
To generate the results and plots for the differential expression analysis between the cell types identified in the Usoskin dataset, run the deAnalysis.Rmd file.
For each of the real datasets Islam, Usoskin, 10X genomics, respectively run benchmark_islam.Rmd, benchmark_usoskin.Rmd, benchmark_tenx.Rmd. Finally, run benchmark_all.Rmd.