Please see https://github.com/pkerrwall/rnaseq_pipeline with updated code, demo, and detailed instructions on setup
for sample in Wt-1 Wt-2 Wt-3 DecWt-1 DecWt-2 DecWt-3 P384R-1 P384R-2 P384R-3; do echo $sample; python trim_galore_pipeline.py $sample; done
cd ~/db/Dmel/dmel_r6.39
extract_splice_sites.py dmel-all-r6.39.gtf > dmel-all-r6.39.gtf.ss
extract_exons.py dmel-all-r6.39.gtf > dmel-all-r6.39.gtf.exon
hisat2-build -p 32 --ss dmel-all-r6.39.gtf.ss --exon dmel-all-r6.39.gtf.exon dmel-all-chromosome-r6.39.fasta dmel-all-chromosome-r6.39.fasta_hisat2_index
cd ~/db/Dmel/dmel_r6.39
~/software/salmon/salmon-1.6.0_linux_x86_64/bin/salmon index -p 32 -t dmel-all-transcript-r6.39.fasta -i dmel-all-transcript-r6.39.fasta_hisat2_index
hisat2_fc_pipeline.py
for sample in Wt-1 Wt-2 Wt-3 DecWt-1 DecWt-2 DecWt-3 P384R-1 P384R-2 P384R-3; do echo $sample; python salmon_aln.py $sample; done
minimap2 arguments from A comprehensive examination of Nanopore native RNA sequencing for characterization of complex transcriptomes (Nature 2019)
p=.80; N=100
threads=32
echo run minimap2 with salmon arguments p=$p
minimap2 -t $threads -ax map-ont -p $p -N $N $transcript_fa $lib/$fq | samtools view -bh > $lib/$fq.transcript_aln.p$p.N$N.bam
samtools sort -@ $threads -o $lib/$fq.transcript_aln.p$p.N$N.sorted.bam $lib/$fq.transcript_aln.p$p.N$N.bam # Sort bam
samtools index -@ $threads $lib/$fq.transcript_aln.p$p.N$N.sorted.bam # Index the sorted bam
samtools view -@ $threads -b -h -F 2308 $lib/$fq.transcript_aln.p$p.N$N.sorted.bam > $lib/$fq.transcript_aln.p$p.N$N.sorted.primary.bam # Create bam with primary alignment only
python salmon_counts.py
python hisat2_counts.py
Sample | Condition |
---|---|
WT_1 | WT |
WT_2 | WT |
WT_3 | WT |
DecWT_1 | DecWT |
DecWT_2 | DecWT |
DecWT_3 | DecWT |
P384R_1 | P384R |
P384R_2 | P384R |
P384R_3 | P384R |
python dge_pipeline.py # need to edit outdir in R script for this location