MUSIC: Model-based Understanding of SIngle-cell CRISPR screening

Introduction of MUSIC

• MUSIC is the first user-friendly topic modelling based pipeline to analyze single-cell CRISPR screening data (independently termed Perturb-Seq, CRISP-seq, or CROP-seq), which could help to prioritize the gene perturbation effect in a cellular heterogeneity level.
• MUSIC is an integrated pipeline for model-based analysis of single cell CRISPR knockout screening data. MUSIC consists of three steps: data preprocessing, model building and perturbation effect prioritizing:
  • Data preprocessing: Besides the conventional quality control and data normalization applied in single-cell RNA-seq analysis, MUSIC addresses two specific considerations that should be taken into account for such novel data type: (1) Filtering perturbed cells with invalid edits; and (2) Filtering perturbations according to a minimal number of cells per perturbation.
  • Model building: MUSIC builds an analytical model based on Topic Models to handle single-cell CRISPR screening data. The concept of topic models was initially presented in machine learning community and has been successfully applied to gene expression data analysis. A key feature of topic model is that it allows each perturbed sample to process a proportion of membership in each functional topic rather than to categorize the sample into a discrete cluster. Such a topic profile, which is derived from large-scale cell-to-cell different perturbed samples, allows for a quantitative description of the biologic function of cells under specific gene perturbation conditions. MUSIC addresses two specific issues when applying the topic model to this specific data type: (1) The distribution of topics between cases and controls is affected by the ratio of their sample numbers, and such a sample imbalance issue is addressed by the bootstrapping strategy when prioritizing the perturbation effect. (2) The optimal topic number is automatically selected by MUSIC in a data-driven manner.
  • Perturbation effect prioritizing: Based on the model-based perturbation analysis, MUSIC can quantitatively estimate and prioritize the individual gene perturbation effect on cell phenotypes from three different perspectives, i.e., prioritizing the gene perturbation effect as an overall perturbation effect, or in a functional topic-specific way and quantifying the relationships between different perturbations.

Introduction of Input File Format

• Input File Format. It should be noted that the starting point of MUSIC analysis is the input of gene expression profile data instead of fastq data. For running MUSIC, the input data needs to follow the standard format we defined. For convenience, MUSIC accepts two kinds of input data formats: (1) The first data format can be referred in the crop_stimulated. You can apply function "Input_preprocess()" to handle this data format. You can format your own data according to the provided example files. (2) The second data format can be referred in the perturb_GSM2396857 generated by 10X genomics. The directory data_format_example/perturb_GSM2396857 contains "barcodes.tsv", "genes.tsv", "matrix.mtx", "cbc_gbc_dict.tsv" and "cbc_gbc_dict_grna.tsv". You can apply function "Input_preprocess_10X()" to handle this data format. For convenient, crop_unstimulated.RData is also an example to show the needed format of MUSIC.
• Attention: （1） The label of the control sample needs to be "CTRL".

Install

• Install: You can install the MUSIC package from Github using devtools packages with R>=3.4.1. For convenience, you can also install the MUSIC package from Docker Hub with the link music

  library(Biostrings)
  library(clusterProfiler)
  library(devtools)
  ## https://github.com/mohuangx/SAVER
  library(SAVER)
  ## If install_github get something wrong like "tar: This does not look like a tar archive", you can run the next code to set curl.
  #options("download.file.method"="libcurl")
  install_github("bm2-lab/MUSIC")
  library(MUSIC)

Tutorial

• For illustration purpose, we took the dataset crop_stimulated as an example. You can load the three files in "crop_stimulated" to R environment.

  expression_profile<-read.table("./crop_stimulated/expression_profile.txt",head=T,row.names=1,sep="\t")
  perturb_information_df<-read.table("./crop_stimulated/perturb_information.txt",head=T,row.names=1,sep="\t")
  perturb_information<-as.character(perturb_information_df[,1])
  names(perturb_information)<-row.names(perturb_information_df)
  # If you don't consider off-target effect, this file is not needed.
  #sgRNA_information_df<-read.table("./crop_stimulated/sgRNA_information.txt",head=T,row.names=1,sep="\t")
  #sgRNA_information<-as.character(sgRNA_information_df[,1])
  #names(sgRNA_information)<-row.names(sgRNA_information_df)
  
  # Have a look at expression_profile
  dim(expression_profile)   

  ## [1] 36722  3259
  

  expression_profile[1:3,1:3]

  ##          TACTTGACCCCN TTACAGCTGAAC CTAAGGCCCTTA
  ## A1BG                0            0            0
  ## A1BG-AS1            0            0            0
  ## A1CF                0            0            0
  

  # perturb_information.
   length(perturb_information)

  ## [1] 3259
  

  • The first step: data preprocessing.

  # For "data_format_example/crop_unstimulated.RData", this function integrates the input data and filters mitochondrial ribosomal protein(^MRP) and ribosomal protein(^RP).
  crop_seq_list<-Input_preprocess(expression_profile,perturb_information)
  
  # For data format like "perturb_GSM2396857" generated by 10X genomics, function "Input_preprocess_10X()" will be suitable. Users can also change this data format to the standard format like "data_format_example/crop_unstimulated.RData", then use function "Input_preprocess()" to process it.
  
  #crop_seq_list<-Input_preprocess_10X("./perturb_GSM2396857")
  

  # cell quality control
  crop_seq_qc<-Cell_qc(crop_seq_list$expression_profile,crop_seq_list$perturb_information,species="Hs",plot=F)
  
  # data imputation （optional）, it may take a little long time without parallel computation.
  set.seed(234)
  crop_seq_imputation<-Data_imputation(crop_seq_qc$expression_profile,crop_seq_qc$perturb_information,cpu_num=30)

  # cell filtering, it may take a little long time without parallel computation.
  crop_seq_filtered<-Cell_filtering(crop_seq_imputation$expression_profile,crop_seq_imputation$perturb_information,cell_num_threshold=20,plot=T,cpu_num=30)

  

  • The second step: model building

  # obtain highly dispersion differentially expressed genes.
  crop_seq_vargene<-Get_high_varGenes(crop_seq_filtered$expression_profile,crop_seq_filtered$perturb_information,plot=T)

  

  # get topics. 
  topic_model_list<-Get_topics(crop_seq_vargene$expression_profile,crop_seq_vargene$perturb_information,topic_number=c(4:6))
  
  # This step may take a long time if you choosed a large scope of topic number. You can run each topic number seperately, then combine them to save time.
  topic_1<-Get_topics(crop_seq_vargene$expression_profile,crop_seq_vargene$perturb_information,topic_number=4)
  topic_2<-Get_topics(crop_seq_vargene$expression_profile,crop_seq_vargene$perturb_information,topic_number=5)
  topic_3<-Get_topics(crop_seq_vargene$expression_profile,crop_seq_vargene$perturb_information,topic_number=6)
  topic_model_list<-list()
  topic_model_list$models<-list()
  topic_model_list$perturb_information<-topic_1$perturb_information
  topic_model_list$models[[1]]<-topic_1$models[[1]]
  topic_model_list$models[[2]]<-topic_2$models[[1]]
  topic_model_list$models[[3]]<-topic_3$models[[1]]
  

  # select the optimal topic number.  
  optimalModel<-Select_topic_number(topic_model_list$models,plot=T)
  
  #If you just calculated one topic number, you can skip this step, just run the following:
  optimalModel<-topic_model_list$models[[1]]

  

  # annotate each topic's functions. For parameter "species", Hs(homo sapiens) or Mm(mus musculus) are available.
  topic_func<-Topic_func_anno(optimalModel,species="Hs",plot=T)

  

  • The third step: perturbation effect prioritizing

  # calculate topic distribution for each cell.
  distri_diff<-Diff_topic_distri(optimalModel,topic_model_list$perturb_information,plot=T)

  

  # calculate the overall perturbation effect ranking list without "offTarget_Info".
  rank_overall_result<-Rank_overall(distri_diff)
  #rank_overall_result<-Rank_overall(distri_diff,offTarget_hash=offTarget_Info) (when "offTarget_Info" is available).
  
  # calculate the topic-specific ranking list.
  rank_topic_specific_result<-Rank_specific(distri_diff)
  
  # calculate the perturbation correlation.
  perturb_cor<-Correlation_perturbation(distri_diff,plot=T)

  

  • If sgRNA sequence of each knockouts were known and you want to investigate if they have off-targets, you can perform this step. This step won't affect the final ranking result, but just present the off-target information. In most cases, the induced sgRNA in such experiment has no off-targets. If you do not want to consider this factor, then just skip this step.

  #library(CRISPRseek)
  #library("BSgenome.Hsapiens.UCSC.hg38")
  #library(TxDb.Hsapiens.UCSC.hg38.knownGene)
  #library(org.Hs.eg.db)
  #gRNAFilePath<-"./crop_stimulated/crop_stimulated_sgrna.fa"
  #crop_results <- offTargetAnalysis(inputFilePath = gRNAFilePath, findgRNAs = FALSE,findgRNAsWithREcutOnly = FALSE,findPairedgRNAOnly = FALSE, BSgenomeName = Hsapiens,txdb = TxDb.Hsapiens.UCSC.hg38.knownGene,min.score=1,scoring.method = "CFDscore",orgAnn = org.Hs.egSYMBOL, max.mismatch = 3,outputDir=getwd(), overwrite = TRUE)
  # then, check if there are off-targets.
  # offTarget_Info<-Get_offtarget(crop_results,crop_seq_filtered$expression_profile,crop_seq_filtered$perturb_information,sgRNA_information)
  

Citation

Duan, B., et al., Model-based understanding of single-cell CRISPR screening. Nat Commun, 2019. 10(1): p. 2233.

Contact

bioinfo_db@163.com or qiliu@tongji.edu.cn