COSEG

Background

A program to identify repeat subfamilies using significant co-segregating ( 2-3 bp ) mutations.

This program is derived from three C programs and several perl scripts written by Alkes Price as part of an analysis of Alu elements in the human genome ( Whole-genome analysis of Alu repeat elements reveals complex evolutionary history, Alkes L. Price, Eleazar Eskin, and Pavel Pevzner, 2004 Genome Research ). The program was first adapted for use with other repeat families and then extended to support consideration of three co-segregating mutations using Alkes statistical model. In 2008 with the help of Andy Siegel an alternative statistical model was developed and the codebase repackaged into the single source file.

There are a few caveats:

• The input sequences must be full length alignments to a single reference sequence.

• The longer the sequence the harder this problem is to solve. Shorter families or short subregions are recommended for this process.

Current Authors: Robert Hubley rhubley@systemsbiology.org Andy Siegel Arian Smit

Original Code Authors: Alkes Price

The original code can be found here: http://www.cse.ucsd.edu/~ppevzner/download/alucode.tar.gz

Installation

1. Build COSEG c program make

2. Edit preprocessAlignments.pl

   Change line 114:

        use lib "/usr/local/RepeatMasker";
   

   To point to your RepeatMasker installation.

3. OPTIONAL: If you plan to run the experimental script refineConsSeqs.pl you will need to edit the script.

   Change line ~110:

        my $RepeatModelerDir = "/usr/local/RepeatModeler";
   

   To point to your RepeatModeler location.

ALU Example Run:

1. Given the sample ALU dataset provided by Alkes Price in the original distribution of his code ( files have been renamed ):

      ALU.seqs: Human sequence data from alignment to AluSx consensus )
      ALU.ins : Insertion sequences from the alignments 
      ALU.cons: AluSx consensus 
                Note: that positions 116-135 (inclusive) are 
                lowercase while all remaining positions are uppercase.
                This is how we encode positions within the consensus
                that have lower quality and should not be considered
                in this analysis.
   

2. Run analysis:

   runcoseg.pl -k -d -m 50 -c ALU.cons -s ALU.seqs -i ALU.ins
   
   NOTE: In this run we must tell coseg to treat lower case
         characters in the consensus as a blacklist designation
         by using the "-d" flag.  Also we are using the 
         original pValue calculation as defined by Alkes et al.
         by specifying the "-k" flag to closely approximate the
         functioning of Alkes original program.
   

3. Create png/svg files

   Coseg outputs a visualization of the subfamily tree in both GraphViz and SVG formats. The SVG format can be viewed directly in any web browser. The GraphViz format needs to be converted to a graphics format ( ie. PNG ) using the GraphViz command line tools ( http://www.graphviz.org/ ) or loaded using the graphViz webapp ( https://mdaines.github.io/viz.js/ ).

   The SVG output may be resized by changing the first line in the SVG file to reflect the output size you would like.

   For example the default is to display the entire graph in the current size of the web browser window:

    <svg height="100%" width="100%" viewbox="0 0 90.625 146.25">
   

   Changing height/width to:

    <svg height="147" width="91" viewbox="0 0 90.625 146.25">
   

   Would display the graph using the full scale. The viewbox values give you the bounding box of the graph at full scale.

Output Files

The following files are named after the aligned sequence file name as a prefix.

*.log -- A log of mutation sites found and the order of the clustering. *.subfamililies.seq -- name, count, P-value and consensus sequence of each subfamily found by our algorithm. (For subfamilies not in the original scaffold, we also include in parentheses the P-value of the scaffold subfamily from which it is derived). *.assign -- for each of the elements, lists the subfamily to which the algorithm has assigned it. *.tree.svg -- evolutionary tree of the subfamilies, in SVG format. *.tree.viz -- evolutionary tree of the subfamilies, in GRAPHVIZ format.

Input Files

*.cons -- Consensus File: The consensus used to generate the aligned seqeunce file. This is in fasta format.

*.seqs -- Aligned Sequences File: Near full length aligned sequences to the consensus file ( one per line ). Deletions are represented by 1 or more "-" characters. Insertions of any length are represented by a single "+" character.

*.ins -- Inserts File: Inserts pulled from the alignments are stored in this file. The aligned sequence file ordering is used in this file. Each insert is encoded as:

        [<position>:<sequence>][<position>:<sequence>]..

ie.
         3:AA 10:AGTTTA

A script is provided to build these files automatically given either wu-blast or cross_match alignment files -- see below.

Running using your own data

1. Cross_match a reference sequence against a genome or database.

   cross_match line1copies consensus -M 25p41g.matrix -gap_init -25 -gap_ext -5 -minscore 200 -minmatch 6 -alignments -bandwidth 50 -word_raw > LINE1

   The example file LINE1 included in this distribution was created using the command line above and can be used directly in the following steps.

2. Determine consensus range to use for analysis ( ie. 298 - 797 bp )

3. Create input files to alkes programs:

   preprocessAlignments.pl -maxEdgeGap 10 -minConsRange 298 -maxConsRange 797 -alignments LINE1

   This will create 3 new files: LINE1.seqs LINE1.ins LINE1.cons

   NOTE: Use the -w flag to preprocessAlignments if you use WUBlast to perform the alignments.

4. Run analysis:

   runcoseg.pl -t -m 50 -c LINE1.cons -s LINE1.seqs -i LINE1.ins

   NOTE: In this examplewe use the "-t" flag to indicate we want to use 3 bp co-segregating mutations as well as 2bp co-segregating mutations when developing subfamilies. Also we use the new pValue calculation ( default ) by leaving out the "-k" flag.

5. Open up a web browser and point it at the file LINE1.seqs.tree.svg. Most browsers support zooming in on svg files. If you want to render the SVG file larger by default simply edit the *.svg file and change the line:

    <svg height="100%" width="100%" viewbox="0 0 90.625 146.25">
   

   to reflect a fixed size for the graph. The "viewbox" values give the absolute size of the drawing so a 1:1 scale would be:

    <svg height="145.25" width="90.625" viewbox="0 0 90.625 146.25">
   

6. If you would like to produce a SVG file without node label or using either the divergence, P-value or subfamily size as the label, simply rerun ./postprocess.pl without the "-l" flag or with "-l div", "-l pv", "-l c" accordingly.

Utility Programs

refineConsSeqs.pl

Coseg uses a rather crude method of building consensus sequences for each subfamily it finds. This script makes use of the refiner script in the RepeatModeler package to build and refine consensus sequences based on subfamily members assigned by coseg.

Usage:

1. From the directory where your coseg results can be found run:

    refineConsSeqs.pl -subConsFile mycosegrun.subfamilies.seq
   

   Where "mycosegrun" is the prefix of the coseg run.

extractSubSeqs.pl

A utility script to extract the sequences for a given coseg subfamily. This uses the coseg input sequence file along with the *.assign file to determine which sequences belong to the queried subfamily number.

Usage:

Version History

In development:

• IUB codes in input sequences caused the code to segfault. Coseg will now randomly choose a nucleotide to substitute each time it encounters one in in the input sequence. It will also inform the user when doing so. Thanks to David Ray for reporting this and suggesting the fix.
• Fixed the svg tag so that the files will directly load in HTML5 web browsers.
• Calculation of divergence has been improved. We now use kimura substition distance with CpG site accounting modifications instead of the mixed substition and indel calculation.

0.2.3:

• Update to Makefile to support newer toolchains

0.2.2:

• Create a *.svg graph file without the need to download/use GraphViz. The layout is handled by an adaptation of algorithm developed by Atze van der Ploeg. The SVG file produced supports various labeling options and subfamily details displayed when a node is hovered over.
• Changed the default colormap for the graph output. Now warm colors denote more diverged subfamilies in the tree while cooler colors represent younger subfamilies. To restore the original color scheme use the new "-o" flag to coseg.
• Added parameter to control the minimum distance between diagnostic sites. Now the user can override the historic value of 10 using the -u flag.
• Improved error reporting when there is a mismatch between an individual sequence length and the consensus length in the input files.
• Fixed a bug that caused coseg to segfault.
• Added experimental script refineConsSeqs.pl. This script uses the RepeatModeler application to build and refine the consensus sequences for each subfamily.

0.2.1:

• Improved code documentation
• Single mutation significance cutoff ( SIGMATHRESH ) was pre-calculated for Alkes Alu analysis and hardcoded. This version calculates the correct sigma cutoff using the length of the input sequence.
• Fixed bug with implementation of Siegel's pValue calculation which caused a segfault -- found by Neal Platt.
• Switched default pvalue method to Andy Siegel's method and provided a new "-k" switch to use Alkes Price's method.
• Fixed bug where the program was exiting when calculations fell below the precision of the machine ( epsilon ). Message given was "Below epsilon..." and the runcoseg.pl script moved on even though coseg failed.