Kallisto

Inroduction

It's the new (2015) way of evaluating gene expression abundance from NGS short reads. It stems from concerns about the, by now, widely used alignment-to-reference step. It introduces extra complexity, and requires closer quality control to avoid erros and effects (Duplicates being one of the unwanted effects). Pachterlab, originator of this sftware felt that this step can be called approximate alignment.

A new program appeared call sailfish which, rather than reads, focused on accurate alignment of the kmers reesulting from those reads. The speed up resulting from skipping the alignment step is considerable.

It is considerably faster than other methods (like those based on say, RSEM) in that it omits the conventional alignment step, and instead calculates what it calls compatibility classes for each read, which are transcripts that the read could align with, if a proper alignment had taken place.

Steps in Brief

First off, we need an assembly of some sort: a reference transcriptome or genome, which may have been de-novo assembled. As is often the case, this needs to be indexed first. Kallisto has its own tool for that. Here we use the example data from the Edgen RNAseq pipeline:

kallisto index -i mm10_chr19-1-20000000.idx mm10_chr19-1-20000000.fasta 

Explanation:

• -i is not the input option but rather the index name option, which is the command is the chosen output name for the index file.
• the reference or assembly follows with no associated option

Armed with the index file, kallisto is now ready to quantify. Here is the basic format of the command:

kallisto quant -i <index_file> -o outputdir <one_pair_of_read_files>

wih several

Explanation:

• Kallisto only takes one sample at a time, which most often will mean one pair of fastq files. If the sample is spread over several pairs of files, they can all follow each other n the command line. This might make it look that kallisto takes more than one sample, but that's wrong. In summary: one sample only to kallisto. The integration of the different samples into the analysis appears to be the job of sleuth.

However, for cluster usage, and for more than a few samples, it is best to use a job script, and specifically, the job-array mode of Gridengine, the queue manager. First the sample filenames should be listed out in a text file, one pair per line. For example, like so:

Sleuth

Sleuth is an associated program for Kallisto, when dealing with several samples for which we have pair-end read sets. It is implemented in R and is available in Bioconductor.

If already installed, it can be loaded, like all other R modules, it - and its dependencies - are loaded as follows:

library(sleuth)

If it is not installed, then here is the method for installing it:

source("http://bioconductor.org/biocLite.R")
biocLite("devtools")    # only if devtools not yet installed
biocLite("pachterlab/sleuth")

Sleuth is made up of the various programs:

• sleuth_prep, preparation stage.
• sleuth_fit, to fit a model
• sleuth_wt, for hypothesis testing
• sleuth_lrt, also for hypothesis testing

see links for more details

Links

• Kallisto's own getting started page at starting
• sleuth manual
• Getting started with Sleuth
• benchtobioinformatics
• Andrew MacKenzie