Our typical RNA quantification pipeline

Transcription

1 RNA-Seq primer

2 Our typical RNA quantification pipeline Upload your sequence data (fastq) Align to the ribosome (Bow>e) Align remaining reads to genome (TopHat) or transcriptome (RSEM) Make report of quality metrics Output ribosomal contamina>on metrics report Produce RNA- Seq report % aligned, % intergenic, % exonic, % UTR Produce IGV/UCSC friendly files Quan>fy transcriptome Call differen>ally expressed genes (if mul>ple samples) Produce a table with normalized expression values Report pairwise significant genes that are differen>ally expressed

3 Initial analysis

4 Summary I Data types, file formats and utilities Annotation: Genomic regions Genes Peaks Bedtools to manipulate them Alignment: Map reads BAM/SAM Samtools to manipulate them Aggregation: Summary files Wig (UCSC) TDF (IGV)

5 Summary II Data process Short read alignment (Bowtie, BWA) Making the genome searchable: Hashing/BW Seed an extend (hashing) vs suffix searches (BW) New aligners are mix Spliced aligners (TopHat, STAR, GSNAP) Map read fragments then strung them Choosing the fragment size Avoiding biases using information (junctions) Quantifying (RSEM/Cufflinks) Read/Isoform assignment Normalization procedures Differential expression (DESeq/EdgeR/Cufflinks)

6 Visualization tricks & Tips Viewing normalized data Downsampling reads to avoid crashes Gene lists Sessions

7 IGV: Integrative Genomics Viewer A desktop applica>on for the visualization and interac>ve explora>on of genomic data Microarrays Epigenomics RNA- Seq NGS alignments Compara:ve genomics

8 Visualizing read alignments with IGV RNASeq

9 Visualizing read alignments with IGV zooming out RNA Seq t K4me3 ChIP Seq PolII ChIPSeq t Cebeb ChIP Seq

10 Viewing several loci simulateneouly: Gene lists

11 Viewing several loci simultaneously: Gene lists

12 Lets create a list using our small dataset Use the following genes Fgf21 Bcat2 Rasip1 Naa60 Which are all within the dataset we selected.

13 Normalizing tracks You can use simple read depth normaliza>on for comparison of different tracks

14 Configure your alignment display Downsampling reads is cri>cal when loading the full alignments. When you are loading reads, downsampling ensures that regions with high coverage result in IGV running out of memory.

15 Saving sessions Sessions allows you to store a set of desired tracks along with any seyng you want

16 Todays topics Looking at ALL of your data

17 Comparing samples: Scatter plots Scatter plot RPKM Sample 2 RPKM Sample 1 RPKM

18 Raw counts/rpkms are NOT Gaussian Density distribution RPKM Density RPKM

19 ... they are more like Log-Gaussian Density distribution log RPKM Density LOG RPKM

20 And log counts/rpkm can be scatter-plotted Control 1e+05 1e+03 1e+01 1e+01 1e+03 Treat 1e+05

21 Which can also be looked at as an MA-Plot Log2 Fold Change (M) Log2 mean normalized counts (A) 30 40

22 Hierarchical clustering vector similarity? Gene Cond1 Cond2 Cond3 Cond4 g g g g Clustering is about similarity: Between two rows (specified by a distance func>on) Between two sets of rows (specified by the linkage method)

23 Common similarity approaches Distance between rows (or columns) Correlation: d(r,s) = (1- cor(r,s))/2 Euclidean: d(r,s) = sqrt(σ i (r i s i ) 2 ) Linkage: Distance betwee two sets (d(r,s)) Complete: Average: Single: G max {d(r, s),s2 S, r 2 R} { 2 2 } mean {d(r,{ s),s22 S, r 2 R} } min {d(r, s),s2 2 S, r 2 R} } Gene Cond1 Cond2 Cond3 Cond4 g g g g

24 The effect of the linkage method Complete linkage correlation Single linkage correlation g1 g2 g1 g2 Height g3 g4 Height g3 g4 Gene Cond1 Cond2 Cond3 Cond4 g g g g

25 Effect of the distance! Complete linkage correlation Complete linkage euclidean g1 g2 g1 g4 Height g3 g4 Height g2 g3 Gene Cond1 Cond2 Cond3 Cond4 g g g g

26 Playing with clustering #Define the toy matrix# ####################### m = rbind (c(2.5,5,7.5,10), c(0.2,0.5,0.8,1.1), c(0.2,0.3,0.4,11), c(2.5,8,8,9)) #Give column and row names# ########################### rownames(m) = c("g1","g2","g3","g4"); colnames(m) = c("c1","c2","c3","c4"); #Compute the correlation distance matrix# ######################################### submat.dist = as.dist( (1 - cor(t(m)) ) /2 ); #Plot clustering with the three main methods# ############################################# plot( hclust(submat.dist, method="complete",members=null), main="complete linkeage - correlation", sub="", xlab="", lwd=3); plot( hclust(submat.dist, method="average",members=null), main = "Average Linkeage - correlation", sub="", xlab="", lwd=3); plot( hclust(submat.dist, method="single",members=null), main = "Single Linkeage- correlation", sub="", xlab="", lwd=3); #Plot clustering with the three main methods, using the euclidean distance# ########################################################################### plot( hclust(dist(m), method="complete",members=null), main="complete linkeage - euclidean", sub="", xlab="", lwd=3); plot( hclust(dist(m), method="average",members=null), main = "Average Linkeage - euclidean", sub="", xlab="", lwd=3); plot( hclust(dist(m), method="single",members=null), main = "Single Linkeage - euclidean", sub="", xlab="", lwd=3);