Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

Version 1 Next »

Raw reads

We received four files with sequence reads. Two of these contain the 1x100bp reads, because two lanes were used on the instrument. Two of these because CU unnecessarily ran indexing reads on the fragments. I deleted these nonsense files. The two files with the raw reads of interest are (these are in /project/microbiome/data/seq/HMAX1/rawreads).

  1. WyomingPool_L1_S1_L001_R1_001.fastq.gz (22 GB) – 457,726,974 reads (109 GBytes uncompressed)

  2. WyomingPool_L2_S2_L002_R1_001.fastq.gz (22 GB) – 450,678,667 reads (107 GBytes uncompressed)

I used unpigz.sh to decompress the fastq files, because our parser does not read from gzipped files.

Demultiplexing

In /project/microbiome/analyses/gtl/HMAX1 I removed extraneous spaces in the file that maps MIDS to individual identifiers (Hmax1Demux.csv). Also, the original Hmax1Demux.csv didn’t follow the scheme we have used for GBS: MIDname, MID, sample id. So, I made a fixed version (now we have Hmax1Demux_fixed.csv): 

sed -E 's/^([[:alnum:]-]+),([[:alnum:]-]+),([[:alnum:]-]+).*/\3,\1,\2/' Hmax1Demux.csv > Hmax1Demux_fixed.csv

This demux key indicated that there were several samples that were technical replicates, but the user also reported one error, which we corrected by hand (see diff below) and made Hmax1Demux_fixed2.csv (this file also lacks a final newline at the end of the file, but I verified this is not a problem

196c196
< C01,CGATATAG,MN-ABR-BLA-9-L
---
> C01,CGATATAG,MN-ABR-BLA-10-L

Demultiplexing on the two files in parallel took more than the two days I initially allocated to it (in part because of the ~10% of the data that do not match our MIDS, because we did not filter contaminants). So I broke the data into 228 parts (each with 16 million lines) and ran 228 jobs in parallel. I repeated this when we learned of the one error in the demux key.

mkdir /gscratch/buerkle/data/HMAX1
cd /gscratch/buerkle/data/HMAX1
cat /project/microbiome/data/seq/HMAX1/rawreads/WyomingPool* | split -l 16000000 -d --suffix-length=3 --additional-suffix=.fastq - WyomingPool_HMAX1_
mkdir rawreads
mv WyomingPool_HMAX1_* rawreads/
/project/microbiome/analyses/gtl/HMAX1/demultiplex/run_parsebarcodes_onSplitInput.pl

Note that I did not do separate contaminant filtering (which I did for Penstemon), because the parsing code and other downstream steps should knock out contaminants. I can double-check this.

I modified the script from 16S/ITS work for splitting fastq files based on information in their info line, to different files. It is: /project/microbiome/analyses/gtl/HMAX1/demultiplex/splitFastq_manyInputfiles_gbs.pl and is run with run_splitFastq_gbs.sh, in the same directory. Output was initially in /project/microbiome/analyses/gtl/HMAX1/demultiplex/sample_fastq. All of this now is in /project/microbiome/data/seq/HMAX1/demultiplex, so that it is reachable thru globus at /project/microbiome/data/seq/HMAX1/.

Eight individuals were duplicated, with different MIDs. Was this planned? I didn’t account for this in the parsing script (the info line only has the individual sample ID, not the MID. I could add it back in. But then the replicates would need to be merged. As is now, all reads for an individual are going into one file. There are also four tubes labeled ‘BLANK' that will all have been merged (all the reads went into BLANK.GGATCCTT.fq).

  • compressed all sample_fastq/ files with pigz: using sbatch /project/microbiome/data/seq/HMAX1/demultiplex/run_pigz.sh

  • moved fastq for all four blank samples (data are all in one file because names are collapsed; noted above) to a subfolder (/project/microbiome/data/seq/HMAX1/demultiplex/sample_fastq/blanks), to get them out of the way.

  • started denovo assembly in /gscratch/buerkle/data/HMAX1/denovo Completed first step for dDocent and am running cd-hit for 92%, 96% and 98% minimum match. Initially didn’t give these enough wall time and in reruns I bumped up the number of cores to 16.

To do:

  • Summarize the parse report files in /gscratch with some code to iterate over all the individual reports and get an overall count.

  • variant calling

  • No labels