1. Why sequence the transcriptome

2. Overview of RNA-seq (From RNA to sequence)

3. RNA-seq analysis (From sequence to RNA and expression)
   • Quality control of sequence data
   • Mapping based approach
   • Transcriptome assembly
   • Estimate gene expression

• Identify gene sequences in genomes
• Investigate patterns of gene expression
• Study isoforms and allelic expression
• Learn about gene functions
• Identify co-expressed gene networks
• Identify SNPs in genes

• Dynamic expression, which means that each tissue, time-point and cell will be different
• Most data comes from "functional units" eg. protein coding genes
• Smaller sequence space as it is a subset of the complete genome

• Experimental design (biology, medicine, statistics)
• RNA Extraction (biology, biotechnology)
• Library preparation (biology, biotechnology)
• Sequencing (engineering, chemistry, physics, biotechnology and bioinformatics)
• Data processing (bioinformatics)
• Data analysis (bioinformatics, biostatistics, biology, medicine)

• The actual sequence data is the same as for genome sequence data
• Both single- and pair-end data
• Major difference is that transcripts often comes from one strand and most data retain this structure eg. the sequence data is from only one strand.

Based on the QC results there are at times a need to filter "bad" data. For example:

• filter reads based on quality score, e.g. with avg. quality below 20 within 4-basepair wide sliding windows
• filter reads that after this are shorter than a given cut-off
• remove unwanted sequences, e.g. adapter sequences or contaminants

• fastQC (QC)
• Prinseq (QC and filtering)
• FastX-toolkit (QC and trimming/filtering)
• R, via ShortRead package (QC and trimming/filtering)
  
• Trimmomatic (Trimming/filtering)
• Cutadapt (Trimming/filtering)

• Read data (fastq files)
  From the sequencing machines

• Reference genome (fasta file)
  The genome sequence of the species.

• Reference annotation (GTF/GFF files)
  Where the genes located on the genome

Available tools: Qualimap, RseQC, Picard, RNA-SeQC

• No reference genome available
• Identify novel transcripts (genes or splice variants)
• Identify fusion genes or transcription from unknown origin

• Assemble the transriptome from the generated short reads
• All useful methods rely on generating a structure called a de-bruijn graph that in this context aims at extenting k-mers in the short reads to longer unambiguous sequences corresponding to transcripts

• SOAP-denovo
• TRANS Oases
• Trans-ABYSS

• How large fractino of reads can be mapped back to the assembly
• Total sequence length of all assembled sequences
• Number of trancripts generated
• Compare to related species with genome information
• Predict open reading frames and protein domains
• etc etc

• Read counts is a measure of expression level
• Many reads mapping to a genomic feature = high expression
• The expression can be quantified on any level eg. gene, transcript (isoforms), or exons.

• Map reads to the assembled transcriptome
• Count reads mapping to the different transcripts that have been assembled.
• It is not as easy as it sound as transcripts can be very redundant and a read might map to multiple transcripts equally well.
• Tools available RSEM, (kallisto and salmon)

• The counts depend on both the number of sequences generated from a samples and on the sequence length of the gene/transcript
• Counts per million (CPM) is used to take out the effect of sequence depth
• Reads per kilobasepair and million (RPKM, FPKM) also takes the lengh of genes into account
• Several other normalisations are available and is used to optimize power in detection of differential expression

• Read counts (and CPM etc) does not follow a normal distribution
• In most experiments few replicates are used, limited power unless we use model based analysis for detection of DE

• It is hence recommended to use softwares designed specifically for RNA-seq DE analysis

• Some useful tools: Cuffdiff (avoid), edgeR, DESeq2, limma

• Many different expertises needed for RNA-seq experiments
• Think ahead, plan wisely, ask for help
• If your experimental design is wrong nothing will help
• Assess and try to improve the quality of raw reads use QC tools and talk to sequencing centre

• Map to genome and use available annotations
• Count reads and check that the majority map according to expectiations
• Does count summaries make sense?
• Use appropriate tools and software for detection of DE

• Use reference genome to guide it, if available
• Spend lots of time in assessing the results e.g. by comparing related species, looking at ORFs etc
• consider merging with other data sources
• Try multiple assemblers

Ensure that your experimental design allows addressing the question of interest. More replicates translates into more power for differential gene expression and easier publication process

• checking the quality of the raw reads with FastQC
• map reads to the reference genome using STAR
• converting between SAM and BAM format using Samtools
• assess the post-alignment read quality using QualiMap
• count reads overlapping with gene regions using featureCounts
• Identify differentially expressed genes using a prepared r-script relying on edgeR

• functional annotation, putting DE genes in the biological context
• exon usage, studying the alternative splicing
• data visualisation and graphics
• de novo transcriptome assembly