Data wrangling best practices

Data wrangling best practices

NGS is smack dab in the middle of the Big Data revolution. Initial NGS FASTQ files are big (100s of MB to GB) – and they're just the start.

Organization and good practices are critical! Your data can get out of hand very quickly!

Keep FASTQ files compressed

• Most sequencing facilities will give you compressed sequencing data files
  • gzip format (.gz extension) for individual files
  • tar or zip format for directories of files
• Even with compression it's easy to run out of storage space!

You may be tempted decompress your sequencing files to manipulate them more directly

• Resist the temptation to gunzip!
• Nearly all modern bioinformatics tools are able to work on .gz files
• There are techniques for working with compressed files without ever decompressing them

Arrange adequate storage space

• At TACC
  • Obtain an allocation on TACC's corral disk array (initial 5 TB are no-cost)
  • Stage your active projects on corral or $WORK
    • copy data to $SCRATCH for analysis
    • copy important analysis products back to corral or $WORK
  • Periodically back up corral or $WORK directories to ranch tape archive
• On a UT Biomedical Research Support Facility (BRCF) "POD"
  • See https://utexas.atlassian.net/wiki/display/RCTFusers/
    • Home and Work areas on POD servers are automatically backed up weekly
      • and are periodically archived to ranch 
  • GSAF customers can obtain a no-cost 2 TB allocation on the shared GSAF POD
    • and use its compute servers
  • PIs can "rent" space on the BRCF Rental pod, and use its 2 compute servers
    • e.g. a 13 TB space allocation for $66/month
    • see BRCF Rental pod offering

Backup analysis artifacts regularly

• All TACC users automatically have a 2 TB allocation TACC's ranch tape archive system
  
  • Larger allocations can be requested by project owners in the TACC User Portal
  • Free!
• Periodically back up your corral or $WORK directories to ranch tape archive
  • Large directories should be combined first using the tar program
    

Distinguish between types of data

Files from different stages of the analysis will have different archival requirements.

• Original sequence data (FASTQ files)
  • Must be backed up!
• Alignments
  • Usually larger than original FASTQs
  • Can be backed up once stable
• Downstream analysis results
• Reporting files (plots, plotting code)

While a project is active you may want to keep more intermediate results for reference. Many of these can be removed after publication.

Track your analysis steps

Your analyses should be reproducible by others so you need to keep the equivalent of a lab notebook to document your protocols.

• Keep "work files" that detail analysis steps performed
  • here's and Example alignment work file