Skip to content

accaldwell/coffee_analysis

Repository files navigation

coffee_analysis: Supplemental data and code for coffee project.

##Requirements

  • Linux (Arch Linux and Ubuntu Server 13.10 used). OS X may run the needed software, but is untested.
  • A large amount of memory. Approximately 137GB RAM is used during the get.oturep() step, and ~47GB is used during the cluster() step. Amazon EC2 instance type cr1.8xlarge was used for memory intensive steps, which provides 244GB of memory.
  • Trimmomatic (version 0.32 used)
  • Mothur (version 1.32.1 used)
  • Python 2 and 3 (versions 2.7.6 and 3.3.3 used)
  • BioPython (version 1.63 used)
  • R (version 3.0.2 used)
  • R libraries ggplot2, plyr, gridExtra
  • Krona

##Obtaining raw reads Raw reads are deposited in Zenodo (http://dx.doi.org/10.5281/zenodo.11120).

DOI

##Cleaning raw reads Prior to running mothur, raw fastq sequences need to be cleaned with trimmomatic. A convenience script is provided at scripts/print_trimm_cmd.py if changes to path names need to be made. By default, the scripts assume the raw reads are in fastqs_raw/

$bash scripts/trimm_raw_reads.sh

##Running mothur

$mothur protocol.batch

Annotations are provided in the mothur batch file, explaining steps. The mothur protocol also runs accessory scripts, with additional requirements as seen above.

##A note on sample naming Sample and sequences have abbreviated names during data analysis.

ICfXX_YY: INT-Coffee, Sample XX, Sequence YY (only for seqences)

ICn: INT-Control

TCf: TRN-Coffee

TCn: TRN-Control

OCf: ORG-Coffee

OCn: ORG-Control

###Contact info primary email: [email protected]