Alignment Free Homology Detection

These are a bunch of scripts that help detect sequence homology without the use of alignment. These mainly align a given training set of regions from hg18 to danRer5 identified using conserved flanks even though the middle region cannot be aligned.

Requirements

• Python 2.7, or 2.6 + argparse
• Perl
• Bash
• Biopython
• numpy, scipy, scikits.statsmodels, matplotlib
• twoBitToFa (UCSC Kent)

Pipeline

To get the 2bit genomes for hg18 and danRer5 (over 1GB):

./get_data.sh

Obtain the formatted file used for most utilities here:

./extract_sequences.sh

Perform the alignment for D2z and HexMCD (takes a long time):

cat hg18.toDanRer5.seqs.txt | ./af.py d2z
cat hg18.toDanRer5.seqs.txt | ./af.py hexmcd

These produce pickled files, which can be converted to .dat as below:

./pkl_to_dat d2z.pkl > d2z.dat

For the alignments of HexDiff, HexYMF:

cat hg18.toDanRer5.seqs.txt | ./hexdiff.pl region 1.0  > hexdiff.dat
cat hg18.toDanRer5.seqs.txt | hexymf/hexymf.pl > hexymf.dat

Use scoring to generate ranked score reports using either the overlap or ranked_peaks algorithm:

./scoring.py -f d2z.dat ranked_peaks > d2z_overlap

Generate cdf plots:

./scoring_cdf.py -f d2z_overlap hexmcd_overlap

You can also plot an individual cne's impulse function:

./plot.py -f d2z.dat <cne_name>