-
Notifications
You must be signed in to change notification settings - Fork 6
Getting started
If you are going to be doing anything computationally intensive (which you probably are), you should put the SmileTrain code on your compute cluster (probably coyote).
Right now the Smile Train is not bundled as a package or anything, so you'll have to download the scripts and put them in the right place. I personally recommend ~/lib/SmileTrain.
For example, if you are placing into ~/lib, you should cd ~/lib and then git clone https://github.com/almlab/SmileTrain.git. (You can also get that https address from the main repository webpage.)
Before trying to run any of the scripts, you need to edit user.cfg. This file tells the SmileTrain scripts where to place temporary job submission scripts, where to look for the other scripts, etc.
You will definitely need to change the username, tmp_directory, library, and bashrc lines. (Make sure the tmp_directory folder exists!) The queue you pick will depend on your needs. (You can learn about the queues on your compute cluster with the obscure command qmgr -c 'p s' or the less informative qstat -Q.) You can point to my usearch, or you can download your own copy.
The cluster and [Data] lines are set up for use on coyote. If you are using a different cluster, you'll have to adjust those lines.
You'll need forward and reverse reads (I'll call them for.fastq and rev.fastq) in Illumina 1.3-1.7 format, a barcode file (I'll call it barcode.txt; it should have lines with sample name and barcode separated by a tab), and the forward and reverse primer (I'll call them AAA and TTT).
If you want to go all the way from your raw data to a reference-based OTU table using Greengenes, you'll just need to run /path/to/SmileTrain/otu_caller.py -f for.fastq -r rev.fastq -p AAA -q TTT -b barcode.txt --all -n 10.
The --all is a shortcut for --check --split --convert --primers --merge --demultiplex --qfilter --dereplicate --index --ref_gg. The -n 10 means that the early steps (converting fastq format through quality filtering) will be performed in parallel on 10 nodes. The number of nodes you pick should be decided from a balance between job submission overhead and pure computational time needed.
If you are starting with a QIIME fasta file, you should read How to process a filtered QIIME fasta.