nJSD is a python package for calculating distance between two biological networks instantiated with gene-expression profiles using entropy concept. It was designed to measure intratumor heterogeneity from bulk RNA-sequencing data. Transcriptome-based ITH (tITH) of tumor state was calculated by considering both normal state and ideally heterogeneous state.
pip install njsd
nJSD supports command-line invocation as below:
usage: njsd [-h] -n NETWORK -r REF -q QUERY -o OUTPUT [-t GENESET]
Calculate network-based Jensen-Shannon Divergence.
optional arguments:
-h, --help show this help message and exit
-n NETWORK, --network NETWORK
Pre-defined network
-r REF, --ref REF Reference gene expression profile
-q QUERY, --query QUERY
Query gene expression profile
-o OUTPUT, --output OUTPUT
Output file.
-t GENESET, --geneset GENESET
Gene set list
Note that -t GENESET
option is optional. If -t
option is specified, gene set-specified nJSD and tITH will be computed. Otherwise, njsd
will compute transcriptome-wide nJSD of the two expression profiles and tITH of query gene expression profile.
Network file, which should be given with -n/--network
option must be formatted as below where each line specifies an edge in the network. njsd
will simply ignore the header by skipping a single line, so you may name each column in a human-friendly way:
GeneA GeneB # Header
GeneSymbol1 GeneSymbol2
GeneSymbol1 GeneSymbol3
GeneSymbol1 GeneSymbol4
...
Gene expression profile, which should be given with -r/--ref
or -q/--query
option must follow the format below. Again, the header doesn't matter. Note that njsd
will automatically apply log2-transformation to expression values by taking log2(expression + 1), we recommend giving njsd
unnormalized expression values, such as raw FPKM, RPKM or TPM.
GeneSymbol ExpressionValue # Header
GeneA 10
GeneB 20
BeneC 30
...
Gene set list, which should be given with -t/--geneset
option file must have the format below. Please be warned that this file should not have a header. The first column denotes names of each gene set(or group), and the following columns represent the members of each group.
Group1Name GeneA GeneB GeneC ...
Group2Name GeneD GeneE GeneF ...
Group3Name GeneA GeneG GeneH ...
...
Toy data, which represents three different gene expression profiles(Toy.profile1, Toy.profile2, Toy.profile3
) which are instantiation of the template network(Toy.network
), are given in example
directory. Following execution scenarios show how to compute nJSD between the gene expression profiles.
The template network is shown:
And figures below are three gene expression profiles instantiated on top of the template network:
You can compute transcriptome-wide nJSD as below:
$ njsd -n example/Toy.network -r example/Toy.profile1 -q example/Toy.profile2 -o profile2_njsd_all.txt
Output file contains nJSD_NT(Normal-to-Tumor nJSD), nJSD_TA(Tumor-to-maximally Ambiguous state), and tITH values. You can think of Normal as Reference, and Tumor as Query gene expression profiles.
$ cat profile2_njsd_all.txt
nJSD_NT nJSD_TA tITH
0.003935020793376432 0.0068202519228746615 0.36586899255754446
Let's compare nJSDs between profile 1 and 2, and profile 1 and 3.
$ njsd -n example/Toy.network -r example/Toy.profile1 -q example/Toy.profile3 -o profile3_njsd_all.txt
$ cut -f1 profile2_njsd_all.txt profile3_njsd_all.txt
nJSD_NT
0.003935020793376432
nJSD_NT
0.007758064133920005
Indeed, the distance between profile1 and 3 is greater than that of profile 1 and 2.
You can compute gene set-specified nJSD by specifying -t/--geneset
option as below:
$ njsd -n example/Toy.network -r example/Toy.profile1 -q example/Toy.profile2 -t example/Toy.geneset -o output_njsd_gene_set.txt
Each line of the output file contains nJSD_NT, nJSD_TA, tITH values for each gene set.
$ cat output_njsd_gene_set.txt
Gene_set_ID nJSD_NT nJSD_TA tITH
1st_pwy 0.00782194587529338 0.00938496594270829 0.45458162150340947
2nd_pwy 0.0 0.004261233542045538 0.0
3rd_pwy 0.00521463058352892 0.00710205590340923 0.4233793390015275
4th_pwy 0.007908518155920452 0.004261233542045538 0.6498504120870645
5th_pwy 0.006257556700234704 0.008522467084091077 0.42337933900152747
Y. Park, S. Lim, J. Nam, S. Kim, Measuring intratumor heterogeneity by network entropy using RNA-seq data, Scientific Reports (2016)