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 now available on Github.

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



Download

BART 1.0

BART 1.0.1 for Python2 – Full package including data libraries
BART 1.0.1 for Python3 – Full package including data libraries

BART 1.0.1 for Python2 – Source code only (need to download data library separately)
BART 1.0.1 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


Supplementary Data

The union DNaseI hypersensitive sites (UDHS) used in the BART model. (They are NOT required for BART installation.)

hg38 UDHS
mm10 UDHS



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-v1.0.1-py3-full.tar.gz:

$ tar zxf BART-v1.0.1-py3-full.tar.gz
$ cd BART-v1.0.1-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 (recommended)

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., BART1.0.1/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

Positional arguments

{geneset,profile}


bart geneset

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

Usage:

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

Example:

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

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.

Usage:

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

Example:

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

Input 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.

Notes:

The input file for bart profile should be BED or BAM format in either hg38 or mm10.

Bed is a tab-delimited text file that defines the data lines, and the BED file format is described on UCSC genome browser website. For BED format input, the first three columns should be chrom, chromStart, chromEnd, and the 6th column of strand information is required by BART.

BAM is a binary version of Sequence Alignment/Map(SAM) format, and for more information about BAM custom tracks, please click here.

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_bart_results.txt is a ranking list of all TFs, which includes the Wilcoxon statistic score, Wilcoxon p value, standard Wilcoxon statistic score (zscore), maximum ROC-AUC score and rank score (relative rank of z score, p value and max auc) for each TF. The most functional TFs of input data are ranked first. The file should be like this:

    TF       statistic  pvalue     zscore    max_auc   rela_rank
    AR       18.654     1.172e-77  3.024     0.954     0.004
    FOXA1    13.272     3.346e-40  2.847     0.924     0.008
    SUMO2    5.213      1.854e-07  3.494     0.749     0.021
    PIAS1    3.987      6.679e-05  2.802     0.872     0.025
    HOXB13   3.800      1.446e-04  2.632     0.909     0.027
    GATA3    5.800      6.633e-09  2.549     0.769     0.028
    NR3C1    4.500      6.789e-06  2.042     0.871     0.040
    GATA6    4.240      2.237e-05  2.602     0.632     0.048
    ESR1     12.178     4.057e-34  1.956     0.700     0.049
    CEBPB    5.265      1.404e-07  2.287     0.602     0.057
    ATF4     3.216      1.302e-03  2.348     0.658     0.065
    TOP1     2.254      2.421e-02  3.057     0.779     0.065

  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_bart_results.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 sign up to BART users Google Group for update announcements and discussions.


Citation

If you use BART in your data analysis, please cite:

BART: a transcription factor prediction tool with query gene sets or epigenomic profiles
Zhenjia Wang, Mete Civelek, Clint Miller, Nathan Sheffield, Michael J. Guertin, Chongzhi Zang
Bioinformatics, doi:10.1093/bioinformatics/bty194 (2018)

If you use "geneset" mode, please also cite:

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 2, 2018