CENTIPEDE fits a bayesian hierarchical mixture model to learn TF-specific distribution of experimental data on a particular cell-type for a set of candidate binding sites described by a motif.

This is a practical tutorial for running CENTIPEDE with DNase-Seq data. It explains how to prepare the data and how to run the analysis. The goal is to predict if a putative transcription factor binding site is actually bound or not. For details about the statistical models underlying the methods, please see (Pique-Regi, et al. 2011).

Read the tutorial online or download the PDF:

• http://slowkow.github.io/CENTIPEDE.tutorial
• centipede-tutorial.pdf

This repository has functions to ease the use of CENTIPEDE:

• centipede_data() converts data to the format required for CENTIPEDE.
• parse_region() parses a string like "chr1:123-456".
• read_bedGraph() reads a bedGraph file with 4 columns: chrom, start, end, score.
• read_fimo() reads a text file output by FIMO and selects sites that meet a significance threshold.

I also provide example data that you can use to follow the tutorial:

• cen is a list with two items:
  • cen$mat is a matrix of read-start counts for 3,337 genomic regions.
  • cen$regions is a dataframe describing those regions.
• site_cons is a vector with mean conservation scores for the 3,337 regions, computed across 100 vertebrates.

Install CENTIPEDE by running this in your shell (not within an R session):

wget http://download.r-forge.r-project.org/src/contrib/CENTIPEDE_1.2.tar.gz
R CMD INSTALL CENTIPEDE_1.2.tar.gz

Next, install the tutorial package:

# This command didn't work for me.
# install.packages("CENTIPEDE", repos="http://R-Forge.R-project.org")

install.packages("devtools")
devtools::install_github("slowkow/CENTIPEDE.tutorial")