DefaultParameters

The following is a list of parameters that can be changed via the command line. However, unless you know what you're doing, it's best to leave these unchanged and stick to the instructions in the Tutorial.

ganesh task

Parameter	Default value	Description
lambda	0.1	normal-gamma prior parameter
alpha	0.1	normal-gamma prior parameter
beta	0.1	normal-gamma prior parameter
mu	0.0	normal-gamma prior parameter
init_num_clust	1/2 nb of genes in data matrix	initial number of clusters
num_steps	100	total nb of Gibbs sampling steps
burn_in	50	nb of Gibbs sampling steps before storing the first clustering solution
sample_steps	100	nb of Gibbs sampling steps between storing successive clustering solutions
score_gain	0.0	Bayesian score difference for keeping two condition clusters as separate leaves in the decision tree vs. merging them

• For a detailed definition of the Bayesian score function and the normal-gamma prior, see eq. (1) in the Gibbs sampler paper.
• One Gibbs sampler step is defined as performing (nb of genes) gene reassignment operations followed by (nb of conditions) condition reassignment operations for each current module.
• The default settings mean that one clustering solution is stored at the end the Gibbs sampling procedure.

tight_clusters task

Parameter	Default value	Description
node_clustering	true	choose between node (true) or edge (false) clustering procedures to create the tight clusters
min_weight	0.25	minimum weight for node or edge clustering procedure
min_clust_size	10	minimal cluster size for node or edge clustering procedure
min_clust_score	2	minimal cluster score for node or edge clustering procedure

• Detailed definitions of these parameters are provided in the Supplementary Methods of the Lemon-Tree paper.

regulators task

Parameter	Default value	Description
lambda	0.1	normal-gamma prior parameter
alpha	0.1	normal-gamma prior parameter
beta	0.1	normal-gamma prior parameter
mu	0.0	normal-gamma prior parameter
num_steps	1100	total nb of Gibbs sampling steps
burn_in	100	nb of Gibbs sampling steps before storing the first condition clustering solution
sample_steps	100	nb of Gibbs sampling steps between storing successive condition clustering solutions
score_gain	0.0	Bayesian score difference for keeping two condition clusters as separate leaves in the decision tree vs. merging them
num_reg	10	Number of regulators sampled per leaf node

• The regulators task uses the same Gibbs sampling procedure as the ganesh task to sample condition clustering solutions for each module, but without reassigning genes, hence one Gibbs sampling step is defined here as performing (nb of conditions) condition reassignment operations for each module.
• Regulators are sampled at each decision tree node according to a logistic regression model, see the module network inference paper.
• The default settings mean that 10 condition clustering solutions and associated decision trees are sampled for each module, and 10 regulators are sampled at each decision tree node.

go_annotation task

Parameter	Default value	Description
go_p_value	0.05	GO p-value cutoff level
go_namespace	biological_process	default namespace for enrichment calculations

• For details, see the BiNGO paper.

figures task

Parameter	Default value	Description
use_global_mean	false	use the global mean for the color gradient in the figures
use_regulator_mean	false	use the regulator mean for the color gradient in the figures
map_file	none	name of 2-column txt file mapping IDs used for clustering to IDs used for figures (e.g. to print gene symbols instead of transcript IDs)
cut_level	0	cut the regulation tree at a given depth for the figures (0 means no cutting)
draw_experiment_names	true	include experiment names in the figures
draw_experiment_color	false	use a color code associated to experiments

• The default settings mean that the color gradient is defined separately for each module based on the mean expression level of genes in the module (as opposed to all genes in the data) and that the regulator panel uses the same color gradient as the module genes. When combining different types of regulator data, setting use_regulator_mean to true is recommended.

• If you want to color code the experiments, you should use the -draw_experiment_color followed by a file name. The file contains experiments numbers corresponding to the order of the experiments in the data file, followed by a space or a tab, followed by a color code. Valid color codes are: black, blue, cyan, darkGray, gray, green, yellow, lightGray, magenta, orange, pink, red.