SC RNA-seq analysis piepline

Time-stamp: "2017-11-02 14:14:48"

Install

export PATH="$PATH:$PWD/Drop-seq_tools-1.12/"

Overall steps

1. Preprocessing (From Fastq to UMI exprssion matrix)
2. Generate Gene type heatmap

Detailed steps

Step1: Preprocessing (From Fastq to UMI exprssion matrix)

• Code: Processing_part1_including_intornic_reads.sh
• Input: two fastq files (one for each R1 and one for R2).
• The output:
  • Genes by cells matrix with UMIs
  • UMIs (transcripts) whose barcodes differ by one basepair are merged together when counting umis/gene
• Please See drop-seq tools for explanation of various functions in the actual script

Step2: Downstream Analysis

All functions can be found in post_processing.R. Require:

• Seurat Package in R
• Seaborn for heatmap visualization in python

prepare "seurat object" and QC

• Usage: seurat_object<- process_umi('nameoffile.txt', 'name_you_want_to_use')
• Basic QC metrics include a plot of the distribution of UMIs in all cells and number of genes in all cells

Clustering and visualization (Using TSNE and KNN clustering)

• Usage: process_umi_step2(seurat_object, k for knn clustering, umi cutoff (to control for dobulets))

• Cell selection thresholds:

  1. min_reads: minimal reads (10000 Habib et al. 2017)
  2. min_genes: minimal number of genes per nucleus/cell (200 Habib et al. 2017)
  3. max_reads: Cells that have an abnormally high number of UMI for a given number of genes are considered to be doublets.
  4. The umi cutoff is currently chosen by visualizing the nGene plotted against the nUMI for all cells.

• To estimate k we can run optimize_k(seurat_object) to get out a table. Ideal cluster choice maximizes Dunn & Shilouette Indicies and minimizes the Davies Bouldin index.

• If you would like to perform dimensionality reduction using SVD instead of PCA, you can use the function in dimensionality_reduction.R to output an updated seurat object. Perform this before using optimize_k(seurat_object)

• Usage: seurat_object<-svd(seurat_object)

Identify cell type specific marker genes

• Usage: `identify_clusters(seurat_object, # of clusters, 'name_of_sample')

The name of sample will be used to name the output files, and the number of cells in each cluster will be outputted to the screen. For each cluster a table with a list of signficantly up or downregulated marker genes as well as a table with the average UMI for each gene will be outputted.

Create a heatmap of the cell type specific genes (python script)

• Input: a matrix of the average umi for each gene in each cell type (cluster).
• outputs:
  1. A normalized heat map (z-score)
  2. raw unnormalized heatmap
  3. A csv table showing the order of genes in the normalized and unnormalized heatmap

Create a Violinplot displaying raw UMI distribution of marker genes (python script)

• Input:
  1. A matrix rows(marker genes), columns(cells) filled with raw UMI counts
  2. A matrix rows(cells) and one column with the cluster identity
• outputs:
  1. Violinplots of raw UMI distribution for 8 marker genes You can change the subplot in the python script to accomodate more or less markers genes. plt.subplot(241) means create a 2x4 grid, and place this subplot in the 1st position of that grid