* Check RNA-seq read quality with FastQC
    Version: 0.11.5

* Map reads to genome with Tophat
  * Build genome index for Tophat using PlasmoDB P. berghei ANKA genome version 28
      Version: bowtie2/2.2.9
      Command: bowtie2-build PlasmoDB-28_PbergheiANKA_Genome.fa <bowtie_ref>
  * Map reads with Tophat
      Versions:
        tophat/2.1.1
        bowtie2/2.2.9
        samtools/1.3.1
        boost/1.60.0
      Command: tophat -p 8 --library-type fr-firststrand --max-intron-length 3000 --min-intron-length 4 --output-dir <output_directory> <bowtie_ref> <1.fastq> <2.fastq>

* Check quality of mapping with flagstats
    Version: samtools/1.3.1
    Command: samtools flagstat <accepted_hits.bam>

* Create gene models with Cufflinks.
    Versions:
      cufflinks/2.2.1
      tophat/2.1.1
      bowtie2/2.2.9
    Command: cufflinks -p 3 -o <cufflinks_out_directory> <accepted_hits.bam> --overlap-radius 1 --min-isoform-fraction 0.04 --pre-mrna-fraction 0.1 --library-type fr-firststrand

* Merge gene models with Cuffmerge
    Version: cufflinks/2.2.1
    Command: cuffmerge -p 3 -o <merged_directory> -s <genome.fasta> --min-isoform-fraction 0 <list_of_cufflinks_out>

* Analyse differential expression with limma/voom
    Versions:
      limma/3.28.14
      edgeR/3.4.0
      Rsubread/1.22.2
      R/3.3.1
    Commands:
      fa_targets <- readTargets(file=fa_bam_list)
      fa_celltype <- factor(fa_targets$CellType)
      fa_design <- model.matrix(~fa_celltype)
      fa_fcounts <- featureCounts(files=fa_targets$BamFile,annot.ext=gff_file,isGTFAnnotationFile=TRUE,nthreads=8,isPairedEnd=TRUE,strandSpecific=2)
      fa_isexpr <- rowSums(cpm(fa_fcounts$counts) > 3) >= 3
      fa_x <- fa_fcounts$counts[fa_isexpr,]
      fa_y <- voom(fa_x,fa_design,plot=TRUE)
      plotMDS(fa_y,xlim=c(-5,5))
      fa_fit <- eBayes(lmFit(fa_y,fa_design))
      write.csv(topTable(fa_fit,coef=2,number=Inf,p.value=0.05),file="limma_out_fa")
      write.csv(topTable(fa_fit,coef=2,number=Inf,p.value=1),file="limma_out_fa_full")

* Truncate transcripts to CDS boundaries with GeneGuillotine.
  * Manually inspect reference genome for overlapping genes and remove overlapping fragments. If you run GeneGuillotine first, it will inform you where these are.
    Versions:
      GeneGuillotine/1.0
      Ruby/2.3.3
    Command: geneguillotine.rb -i <merged.gtf> -g <genome_no_overlaps.gff> -o <split_merged.gff>

* Analyse differences in alternative splicing with DEXSeq.
    Versions:
      DEXSeq/1.26.0
      Python/2.7.15
      samtools/1.9
      R/3.5.1
    Commands:
      * Create flattened GFF.
          python2 dexseq_prepare_annotation.py <input.gtf> <flattened.gff>
      * For each bam file,
          samtools view -o <accepted_hits.sam> <accepted_hits.bam>
          sort -k1,1 -k2,2n <accepted_hits.sam> > <accepted_hits.sorted.sam>
          python2 dexseq_count.py --p yes -r name -s reverse <flattened.gff> <accepted_hits.sorted.sam> <outputfile>
      * Import into R, then
          BPPARAM = MulticoreParam(workers=8)
          dxd_comparison_1 = DEXSeqDataSetFromHTSeq(countfiles = file.path(inDir, paste(rownames(samples))), sampleData = samples, design= ~ sample + exon + condition:exon, flattenedfile = annotationfile)
          dxd_comparison_1 <- estimateSizeFactors(dxd_comparison_1)
          dxd_comparison_1 = estimateDispersions(dxd_comparison_1, BPPARAM=BPPARAM)
          dxd_comparison_1 = testForDEU(dxd_comparison_1, BPPARAM=BPPARAM)
          dxd_comparison_1 = estimateExonFoldChanges(dxd_comparison_1, fitExpToVar="condition", BPPARAM=BPPARAM)
          dxr1_comparison_1 = DEXSeqResults(dxd_comparison_1)
          DEXSeqHTML(dxr1_comparison_1, FDR=0.05, color=c("#FF000080", "#0000FF80"))
          write.table(dxr1_comparison_1[order(dxr1_comparison_1$padj),], "/path/to/output_comparison_1.csv", sep="\t")

* Analyse alternative splicing with JunctionJuror.
    Versions:
      JunctionJuror/1.0
      Ruby/2.3.3
    Commands: junctionjuror.rb -j <junctions.bed.list> -g <genome_no_overlaps.gff> -o <output_genes> -t 3