ChiNet:
Comparative chi-square network/pathway analysis
ChiNet determines whether a
network or sub-network, e.g., a pathway, is conserved or differential in either
topology or strength across two conditions. ChiNet compares interactions collectively in
the network based on accumulative heterogeneity and homogeneity over all
interactions. The two networks under comparison are assumed to share the same
set of nodes.
The core of ChiNet is a decomposition
rule. Using parent nodes on the pathway topology as candidates, the active
parents of each child are selected for each condition, doable by various
network inference methods. The software has an option to use best fit to data
under each condition as judged by the smallest p-value of chi-squares computed on
contingency tables separately under each condition. The user can also specify
two network topologies reconstructed by other means and only use ChiNet to
highlight differential and conserved interactions. Then a chi-square
measuring interaction homogeneity is computed for a contingency table pooling
all data for each node. By subtracting the homogeneity chi-square from a total
interaction chi-square (sum of chi-squares under each condition), we obtain a
third chi-square measuring interaction heterogeneity for each node. Summing up
chi-squares of homogeneity and heterogeneity respectively over each node,
ChiNet obtains chi-squares of network heterogeneity, network homogeneity, and
total network activity for the entire network/pathway. The three chi-square
statistics satisfy a decomposition rule central to our analysis: Network Total
Activity = Network Heterogeneity + Network Homogeneity.
The decomposition rule implies that
knowing any two of the components can determine the third. The three statistics
allow one to determine whether a network is conserved or differential across
contexts. When nodes in the networks cannot be assumed to be independent
regardless of whether the networks are differential or conserved, ChiNet offers
a bootstrapped gamma approximation to the above three statistics, in addition
to the chi-square approximation.
The input to ChiNet is a
network topology and observed data of nodes in the network under two
experimental conditions. Prior knowledge on the topology of a general
biological network or pathway is required. ChiNet adapts the general topology
to reflect only active interactions associated with the specific experimental
condition.
Specifically, the input includes two
trajectory collection files corresponding to each condition and one or more
network/pathway topology files. The file formats are described in other
documentation (TrajColFile.pdf)
associated with ChiNet. Continuous values must be first discretized to use
ChiNet. One option is to use R package Ckmeans.1d.dp (Wang and Song, 2011) which employs guaranteed optimal 1-D k-means
clustering.
Using one of the -z or -H options, one
can interface ChiNet with other network inference software, by converting the
output of other network inference program to the require
input network/pathway format for ChiNet.
The output of ChiNet
consists of network homogeneity, heterogeneity, and total activity across all
experimental conditions. Based on the statistical significance of network heterogeneity
and homogeneity, a network/pathway is declared to be either differential or
conserved across conditions. The output also includes two sets of statistics
for total parent and child working zone change of nodes on the given pathway.
The output statistics and associated
statistical significance represented by p-values are saved in a text file.
Three Graphviz DOT files visualizing differential and
conserved interactions and nodes are also produced. These files can be
converted to PDF and other graphical format for visualization using Graphviz tools.
The same set of statistics for each
individual nodes and interactions is printed out on the screen and can be saved
to intermediate files using optional command line arguments.
The ChiNet
binary
executable files by clicking on the following links:
|
Windows 7 |
openSUSE Linux |
MacOS X |
·
ChiNet
User
Guide (ChiNet_User_Guide.pdf)
defines the command line arguments and an example to use the program.
·
Trajectory
Collection File Format
(TrajCollFile.pdf) defines
the input trajectory file format required by ChiNet.
·
Graph
Definition File Format
(GDF.pdf) defines the input pathway
file format required by ChiNet.
Click to
download 1.trj and 2.trj, two trajectory collection files (TCFs) and PWY.txt,
the pathway definition file.
To run the
program under Linux, type the following command in your terminal in the
directory/folder where you have downloaded the executable and the input
files:
./ChiNet-linux
-M comparison -1 1.trj -2 2.trj -H PWY.txt -k 2 -w POOLED -Y BY_EACH_COND -L
1000 -D Summary.dot
You will find
a result file (PWY-PWStats.txt), four graph file of interactions
in the Graphviz DOT format (Summary.dot,PWY-Diff.dot,PWY-Conserved.dot,PWY-WkZn.dot), and the screen output.
How to cite ChiNet
Zhang, Y., Z. L. Liu, and M. Song.
(2015) ChiNet uncovers rewired transcription subnetworks in tolerant yeast for
advanced biofuels conversion, Nucleic Acids Research. (In press)