Samples were taked from two different time points: embryonic (EB) or adult (DE). Some CAGE libraries are technical replicates of each other (T1/T2) because there was enough RNA to make two libraries. Otherwise are biological replicates of each other (EB1/EB2, DE1/DE2). Biological replicates were loosely staged in order to screen a broader set of transcriptional states.
Splice leader
sequences (ACTCATCCCATTTTTGAGTCCGATTTCGATTGTCTAACAG) were recognised with
the TagDust program embedded
in a Nextflow pipeline
(tag v1.0.0 revision ID 2dd9ca3901), producing two sets of read pairs per sample.
nextflow run oist/plessy_splitspliceleaderpe --input input.csv -profile oist -w /flash/LuscombeU/charles-plessy/cache/deletemeInOctober --arch SL.Oikopleura_diocia.arch --rrna OKIrRNA.fa
The SL.Oikopleura_diocia.arch file contains:
tagdust -1 S:ACTCATCCCATTTTTGAGTCCGATTTCGATTGTCTAACAG -2 R:N
tagdust -1 R:N
Files containing reads where the splice leader was found and removed kept
their original name (such as DE1_T1_SL_READ1.fq.gz), and files with the
reads that did not contain the splice leader have _un added to their
name (such as DE1_T1_SL_un_READ1.fq.gz).
I did not use the rRNA-filtered reads produced by TagDust, as it removed large numbers of reads that do not match rDNA well...
We used nf-core RNA-seq pipeline version 3.12 to
align the CAGE reads.
nextflow run nf-core/rnaseq -r 3.12 -name CAGE_Oki_nf_PE_resume1 -profile oist -work-dir /flash/LuscombeU/deletemeCharlesPlessy/nf_CAGE_Oki_20220106_PE -params-file nf-params.json -resume
As a control, I aligned Read 1 (R1) and Read 2 (R2) alone in single-end mode. It showed that alignment rate was increased by trimming R2, but not R1. I decided to remove 70 bp from R2. At this value, multimapping starts to increase and alignment rate and exon rate stop to increase. 5' bias stops to decrease. Error rate does not stop to decrease.
I remove the first 6 bases of Read2 because this is where the primer anneals and this part of the reads has a different base sequence composition. This said, it did not make a practical difference in terms of alignment success in single-end control mapping of R2 alone.
The CAGE libraries were then aligned paired-end with the following parameters:
{
"input": "input.csv",
"email": "[email protected]",
"fasta": "\/bucket\/LuscombeU\/common\/Oikopleura\/Genomes\/OKI2018_I69_1.0\/OKI2018_I69_1.0.fa",
"gtf": "Okinawa.genes.gtf",
"hisat2_build_memory": "50.GB",
"clip_r2": 9,
"three_prime_clip_r2": 30,
"aligner": "hisat2",
"seq_center": "DNAFORM",
"save_unaligned": true,
"skip_preseq": true,
"skip_dupradar": true,
"skip_markduplicates": true,
"remove_ribo_rna": true,
"ribo_database_manifest": "rrna-db.txt"
}
Preseq and duprader were skipped (--skip_preseq, --skip_dupradar) for
paired-end alignment because they take a lot of time. Picard markduplicates
was also skipped (--skip_markduplicates) because there was no PCR
amplification.
The pipeline removes sequences matching the ribosomal DNA (rDNA)
sortmerna tool.
The reference rDNA region is beginning of the the pseudo-autosomal region (PAR)
of chromosome 3, which we deposited in Genbank as
OP113812.
The pipeline runs Trimgalore, which finds few Illumina adapter sequences although a MGI sequencer was used. I tolerate these false positives as we need to trim for length anyway.
I could not convert the resulting alignments to BED12 format because the pairedBAMtoBED12 does not handle the case when R1 and R2 overlap and both align to the same splice junction.
Alignments in BAM format were clustered with the Bioconductor package CAGEr and exported to BED and BED12 formats for upload in the ZENBU genome browser using the zenbu_upload command-line tool. Some notes can be found in the CAGEr.Rmd file.
TO Do:
Check the unaligned reads that BLAST well: do they match regions with aligned reads or do they highlight blind spots ?
Tidy CAGEr.Rmd, and give it a proper Rmarkdown parameter handling.