daler / enhancer-snakemake-demo

Demos a Snakemake workflow to classify enhancer regions based on publicly available chromatin marks.

Overview

This is a Snakefile demo to identify candidate enhancer regions in mouse embryonic liver using data from PMID: 22763441. This uses a different method than the authors used to identify enhancers.

Input is an internet connection (files are downloaded from GEO). Output is 2, 3, 4, and 5-state models from ChromHMM.

Prepare

• Install Snakemake: https://bitbucket.org/johanneskoester/snakemake. Snakemake blends the best of Bash, Python, and Makefiles.
• Download http://compbio.mit.edu/ChromHMM/.
• Edit config.py to point to the paths on disk.
• Prepare the BED files you'd like to check against chromatin states by adding them to the compare/links directory. There are already some data from related ENCODE data (created by running the get-data.py script) and some positive mouse enhancers from enhancer.lbl.gov (from running enhancer.lbl.gov.py). These files are small enough to include in the repo, hence they're not downloaded in the Snakefile.

Run

• Run snakemake -npr as a dry-run to see what will be run.
• Run snakemake -pr -j$N , where $N is the number of CPUs, to run the pipeline.

Output

For each number of states $s, see:

• output/$s-state/webpage_$s.html for the states,
• output/$s-state/*_enrichment.png for the enrichment with supplied BED files.
• output/$s-stats/*_dense.bed for a BED file to upload to UCSC.

For example, state 2 in this 4-state model has strongest emmission parameters for marks we'd expect over enhancers, and has the strongest enrichment for other factors we'd expect to be at/near enhancers.

enhancer-snakemake-demo's Issues

Only getting half the data from enhancer.lbl.gov

It seems as though you are only grabbing the coordinates in the fasta file that are annotated a Mouse. However this is only grabbing about half of the data. The enhancer guys do a provide the coordinates between human and mouse on the interface, but not in the fasta file. I would suggest grabbing human data and doing a lift over to map human to mouse coordinates in order to get all the rested enhances