Skip to main content

Extended bias modelling using "LiBiNorm model"

The "LiBiNorm model" mode allows a slightly more detailed exploration of global bias to be performed than is the case with "LiBiNorm model". It takes as input data a "landscape file" that can be generated by LiBiNorm which contains the locations of the reads within the transcripts. This loads more quickly than the bam and feature files so "LiBiNorm model" is more efficient if different options are being explored.

Generating the landscape file

"LiBiNorm count" should be run with the -u <outputfileroot> and -l which generates a landscape file named <filenameroot>_landscape.txt.

Running LiBiNorm Model

The command line format is LiBiNorm model [options] <landscapefile> where <landscapefile> is the name of the landscape file generated by LiBiNorm count.

The options includes all of the all of the bias normalisation options available in LiBiNorm count. In addition the following bias normalisations are available with LiBiNorm model

-r <runs>, --runs=<runs>

LiBiNorm uses a number of Markov Chain Monte Carlo (MCMC) to determine the parameter uncertainty which can be changed with this option (default 10)

-s <length>, --mcmc=<length>

Allows the length of each MCMC run to be changed (default 200).

-g <N>, --genes=<N>

Sets the maximum number of genes to be used for parameter discovery (default all genes)

-k, --skip

The process of parameter determination uses the Nelder Mead simplex method to locate an initial set of parameters for the second stage which uses MCMC to identify the uncertainty associated with each of the parameters. This option results in the initial Nelder Mead stage being omitted and only the MCMC stage being used to determine the coefficients. When this is done, the parameters for the starting point for each run are determined by a random selection from within the likely range of parameters which is used within LiBiNorm as a prior. When this mode is selected then it is necessary to use more (-r), and longer (-s) runs.