Table 1 Comparison of main features in NATbox and BNArray

Provides a Graphical User Interface (GUI)

Command-line driven with no graphical user interface (GUI).

Supported across operating systems: Windows and Linux.

Supported across operating systems: Windows and Linux.

Parameter recommendations are provided for the various algorithms with default values automatically inserted in GUI.

No parameter recommendations and default values are provided for the algorithms.

Input file is tab-delimited where columns represent the genes of interest and rows represent the experiments. The rows are assumed to be independent of one another.

Input file is tab-delimited where rows represent the genes of interest and columns their measurements across experiments. Unlike NATbox, no specifications are provided as to whether the experiments need to be correlated or uncorrelated.

Imputation of missing values is accomplished by k-nearest neighbour approach impute.knn from the R-package impute

Imputation of missing values is accomplished by local least square estimation LLSImpute also implemented in the R-package pcamethods.

Bayesian Structure Learning (BSL) algorithms are invoked from the R-Package bnlearn.

Bayesian Structure Learning (BSL) algorithms are invoked from the R-Package deal.

BSL algorithms ideally suited for continuous as well as discrete data sets characteristic of gene expression profiles and their quantized/coarse-grained counterparts.

BSL algorithm is ideally suited for mixed data type consisting of continuous as well as discrete values. Although, the authors of BNArray use it for analyzing temporally correlated gene expression data.

BSL algorithms from the package bnlearn include constraint-based (GS, IAMB, Fast-IAMB, IIAMB, MMPC) as well as search and score techniques (Hill-Climbing) are implemented.

No options are provided for multiple BSL algorithms.

BSL techniques implemented from bnlearn provide choice of several conditional independence tests and scoring criteria for continuous and discrete data sets under constraint-based and search and score techniques respectively.

Conditional independence tests for constraint-based: mutual information, mutual information for Gaussian distributed data, fast mutual information, Pearson X², Akaike information criterion, linear correlation and Fisher's Z.

Scoring Criteria for search and score: multinomial likelihood, multinomial log-likelihood, Akaike information criterion, Bayesian information criterion, Bayesian Dirichlet score and Gaussian posterior density).

No options are provided for multiple BSL algorithms.

Provides options to incorporate structural priors during BSL by whitelisting (include) and blacklisting (exclude) edges.

No options are provided for incorporating structural priors.

Confidence of an edge is determined by bootstrapping. Uses R-package RGraphviz for visualization, which is designed to handle the layout of large graphs.

Provides options to highlight edges whose confidence is greater than user specified threshold.

Confidence of an edge is determined by bootstrapping. Uses R-package dynamicGraph for visualization, which may require manual tuning of the node layout for large graphs.

No options are provided to highlight edges whose confidence is greater than user specified threshold.

Parallelization of the bootstrap routines is accomplished by invoking functions from the R-package SNOW.

No options are provided for parallelization.

Topological properties of the results of BSL are investigated using centrality measures (degree, betweenness and closeness) from the R-package igraph.

Does not provide any centrality measures.

Provides motif finder from the package igraph for identifying motifs from the results of BSL. Results from igraph can also be loaded into Cytoscape fro detailed visualization.

Provides a modified version of the algorithm CODENSE for constructing coherent sub-networks from the results of BSL.

Provides a text retrieval interface to retrieve published literature to retrieve functional relationships of interest. This is useful justifying the choice structural priors in BSL.

No interface for text retrieval is provided.