BART

Binding Analysis for Regulation of Transcription



About BART

BART (Binding Analysis for Regulation of Transcription) is a bioinformatics tool for predicting functional transcription factors (TFs) that bind at genomic cis-regulatory regions to regulate gene expression in the human or mouse genomes, given a query gene set or a ChIP-seq dataset as input. BART leverages 3,485 human TF binding profiles and 3,055 mouse TF binding profiles from the public domain (collected in Cistrome Data Browser) to make the prediction.

BART is implemented in Python and distributed as an open-source package along with necessary data libraries.

BART is developed and maintained by the Chongzhi Zang Lab at the University of Virginia.



Download

BART 1.0

BART 1.0 for Python2 – Full package including data libraries
BART 1.0 for Python3 – Full package including data libraries

BART 1.0 for Python2 – Source code only (need to download data library separately)
BART 1.0 for Python3 – Source code only (need to download data library separately)

BART Data Libraries

The human genome (hg38) and the mouse genome (mm10) are supported. Test data are included in each library.

hg38 library
mm10 library



Installation

Prerequisites

BART uses Python's distutils tools for source installation. Before installing BART, please make sure either Python2 (Python2.7 or higher is recommended) or Python3 (Python 3.3 or higher is recommended) is installed in the system, and the following python packages are installed:


Install the full package (All data included, requires at least 30GB hard drive storage in the installation directory)

To install a source distribution of BART, unpack the distribution tarball and open up a command terminal. Go to the directory where you unpacked BART, run the install script to install BART globally or locally. For example, if you want to install the package BART-v1.0-py3-full.tar.gz:

$ tar zxf BART-v1.0-py3-full.tar.gz
$ cd BART-v1.0-py3-full

Install with root/administrator permission (by default, the script will install python library and executable codes globally):

$ python setup.py install

If you want to install everything under your own directory, for example, a directory as /path/to/bart/, use these commands:

$ mkdir -p /path/to/bart/lib/pythonX.Y/site-packages
$ export PYTHONPATH=/path/to/bart/lib/pythonX.Y/site-packages/:$PYTHONPATH
$ python setup.py install --prefix /path/to/bart
$ export PATH=/path/to/bart/bin/:$PATH

In this value, X.Y stands for the major–minor version of Python you are using (such as 2.7 or 3.5 ; you can find this with sys.version[:3] from a Python command line).


Configure environment variables

You’ll need to add those two lines in your bash file (varies on each platform, usually is ~/.bashrc or ~/.bash_profile) so that you can use the BART command line directly:

$ export PYTHONPATH=/path/to/bart/lib/pythonX.Y/site-packages/:$PYTHONPATH
$ export PATH=/path/to/bart/bin/:$PATH


Install from source package without data libraries

You can download the Human or Mouse Data Library separately under your own directory. In this case, you have to edit the config file (e.g. BART1.0/BART/bart.conf) after you unpack the source package to provide the directory for the data. For example, if you download the hg38_library.tar.gz (or mm10_library.tar.gz) and unpack it under /path/to/library, then you can modify the bart.conf file as:

hg38_library_dir = /path/to/library/

Then you can run the install script and install BART source package globally or locally same as the full package described above.



Tutorial

Usage

BART [-h] [--version] {geneset,profile}

bart geneset <-i infile> <-s species> [--outdir dir] [-t target] [-p process] [options]

e.g.: bart geneset -i name_enhancer_prediction.txt -s hg38 --outdir bart_output -t target.txt -p 4

bart profile <-i infile> <-f format> <-s species> [--outdir dir] [-t target] [-p process] [options]

e.g.: bart profile -i ChIP.bed -f bed -s hg38 --outdir bart_output -t target.txt -p 4


bart geneset

Given a query gene set (at least 100 genes recommended), predict functional transcription factors that regulate these genes.

Input arguments:

-i <file>, --infile <file>

Input file, the name_enhancer_prediction.txt profile generated from MARGE.

-s <species>, --species <species>

Species, please choose from "hg38" or "mm10".

-t <target>, --target <target>

Target transcription factors of interests, please put each TF in one line. BART will generate extra plots showing prediction results for each TF.

-p <processes>, --processes <processes>

Number of CPUs BART can use.

--nonorm

Whether or not do the standardization for each TF by all of its Wilcoxon statistic scores in our compendium. If set, BART will not do the normalization. Default: FALSE.

Output arguments:

--outdir <outdir>

If specified, all output files will be written to that directory. Default: the current working directory

-o <ofilename>, --ofilename <ofilename>

Name string of output files. Default: the base name of the input file.

Notes:

The input file for bart geneset, i.e., the enhancer_prediction.txt file generated by MARGE, might have two different formats below (depending on Python versions py2 or py3):

a. Python2 version:

1     98.19
2     99.76
3     99.76
4     9.49
5     44.37
6     18.14

b. Python3 version:

chrom    start         end           UDHSID     Score
chr3     175483637     175483761     643494     3086.50
chr3     175485120     175485170     643497     2999.18
chr3     175484862     175485092     643496     2998.28
chr3     175484804     175484854     643495     2976.27
chr3     175491775     175491825     643507     2879.01
chr3     175478670     175478836     643491     2836.90


bart profile

Given a ChIP-seq data file (bed or bam format mapped reads), predict transcription factors whose binding pattern associates with the input ChIP-seq profile.

Input files arguments:

-i <file>, --infile <file>

Input ChIP-seq bed or bam file.

-f <format>, --format <format>

Specify "bed" or "bam" format.

-n <int>, --fragmentsize <int>

Fragment size of ChIP-seq reads, in bps. Default: 150.

-s <species>, --species <species>

Species, please choose from "hg38" or "mm10".

-t <target>, --target <target>

Target transcription factors of interests, please put each TF in one line. BART will generate extra plots showing prediction results for each TF.

-p <processes>, --processes <processes>

Number of CPUs BART can use.

--nonorm

Whether or not do the standardization for each TF by all of its Wilcoxon statistic scores in our compendium. If set, BART will not do the normalization. Default: FALSE.

Output arguments:

--outdir <outdir>

If specified, all output files will be written to that directory. Default: the current working directory

-o <ofilename>, --ofilename <ofilename>

Name string of output files. Default: the base name of input file.


Output files

  1. name_auc.txt contains the ROC-AUC scores for all TF datasets in human/mouse, we use this score to measure the similarity of TF dataset to cis-regulatory profile, and all TFs are ranked decreasingly by scores. The file should be like this:

    AR_56254     AUC = 0.954
    AR_44331     AUC = 0.950
    AR_44338     AUC = 0.949
    AR_50273     AUC = 0.947
    AR_44314     AUC = 0.945
    AR_44330     AUC = 0.943
    AR_50100     AUC = 0.942
    AR_44315     AUC = 0.942
    AR_50044     AUC = 0.926
    AR_50041     AUC = 0.925
    FOXA1_50274     AUC = 0.924
    AR_50042     AUC = 0.921

  2. name_wilcoxon_test.txt is a ranking list of all TFs, which includes the Wilcoxon statistic score, Wilcoxon p value, standard Wilcoxon statistic score (z score) and rank score (average rank of z score and p value) for each TF. The most functional TFs of input data are ranked first. The file should be like this:

    TF    p_value    wilcoxon_score    z_score    avg_z_p
    AR    1.121e-77    18.656    3.024    2.5
    FOXA1    3.189e-40    13.276    2.847    4.5
    SUMO2    1.746e-07    5.225    3.501    5.5
    GATA3    6.378e-09    5.807    2.551    11.5
    PIAS1    6.657e-05    3.988    2.803    14.0
    GATA6    2.208e-05    4.243    2.604    14.5
    CEBPB    1.347e-07    5.272    2.289    16.5
    HOXB13    1.441e-04    3.801    2.632    17.0
    ESR1    3.640e-34    12.187    1.957    22.5
    NR3C1    6.776e-06    4.501    2.042    24.0
    ATF4    1.294e-03    3.217    2.349    29.5
    GATA4    4.264e-03    2.858    2.449    33.5

  3. name_plot is a folder which contains all the extra plots for the TFs listed in target files (target.txt file in test data). For each TF, we have boxplot, which shows the rank position of this TF in all TFs (derived from the rank score in name_wilcoxon_test.txt), and the cumulative distribution plot, which compares the distribution of ROC-AUC scores from datasets of this TF and the scores of all datasets (derived from the AUC scores in name_auc.txt).


Frequently Asked Questions

Please contact us for any questions!


Citation

BART manuscript is in preparation.

If you use geneset mode, please cite the MARGE paper:

Modeling cis-regulation with a compendium of genome-wide histone H3K27ac profiles
Su Wang*, Chongzhi Zang*, Tengfei Xiao, Jingyu Fan, Shenglin Mei, Qian Qin, Qiu Wu, Xujuan Li, Kexin Xu, Housheng Hansen He, Myles Brown, Clifford A. Meyer, X. Shirley Liu
Genome Research 26, 1417–1429 (2016)


Contact

BART is developed and maintained by the Chongzhi Zang Lab at the University of Virginia. Please email us for any questions.


Last modified: August 15, 2017