view release on metacpan or  search on metacpan

scripts/correlate_position_data.pl  view on Meta::CPAN

=item --norm [rank|sum]

Optionally define a method of normalizing the scores between the 
reference and test data sets prior to calculating the correlation. 
Two methods are currently supported: "rank" converts all values 
to rank values (the mean rank is reported for identical values) 
and essentially calculating a Spearman's rank correlation, while 
"sum" scales all values so that the absolute sums are identical. 
Normalization occurs after missing or zero values are interpolated. 
The default is no normalization.

=item --force_strand

If enabled, a strand orientation will be enforced when determining the 
optimal shift. This does not affect the correlation calculation, only 
the direction of the reported shift. This requires the presence of a 
data column in the input file with strand information. The default is 
no enforcement of strand.

=back

=head2 General options

=over 4

=item --gz

Specify whether (or not) the output file should be compressed with gzip.

=item --version

Print the version number.

=item --help

Display this POD documentation.

=back

=head1 DESCRIPTION

This program will calculate statistics between the positioned scores of
two different datasets over a window from an annotated feature or
chromosomal segment. These statistics will help determine whether the
positions or distribution of scores across the window vary or underwent
a positional shift between a test and a reference dataset. For example,
if the enrichment of nucleosome signal from a ChIP experiment shifts in
genomic position, indicating a change in nucleosome position. 

Two statistics may be calculated. First, it will calculate a a Pearson
linear correlation coefficient (r value) between the datasets (default). 
Additionally, an ANOVA analysis may be performed between the datasets and 
generate a P-value. 

By default, the correlation is determined between the data points 
collected over the entire length of the feature. Alternatively, a 
radius and reference point (default is midpoint) may be provided 
that sets the window for collecting scores and calculating a correlation.

In general, to ensure a more reliable Pearson value, fragment ChIP or 
nucleosome coverage should be used rather than point (start or midpoint) 
data, as it will give more reliable results. Fragment coverage is more 
akin to smoothened data and gives better results than interpolated point 
data. 

Normalized read-depth data should be used when possible. If necessary, 
Values can be normalized using one of two methods. The values may be 
converted to rank positions (compare to Kendall's tau), or scaled such 
that the absolute sum values are equal (for example, when working with 
sequence tag read counts).

In addition to calculating a correlation coefficient, an optimal shift 
may also be calculated. This essentially shifts the data, 1 bp at a time, 
in order to identify a shift that would produce a higher correlation. In 
other words, what amount of movement to the left or right would make the 
test data look like the reference data? The window is shifted from -2 
radius to +2 radius relative to the reference point, and the highest 
correlation is reported along with the shift value that generated it. 

=head1 AUTHOR

 Timothy J. Parnell, PhD
 Dept of Oncological Sciences
 Huntsman Cancer Institute
 University of Utah
 Salt Lake City, UT, 84112

This package is free software; you can redistribute it and/or modify
it under the terms of the Artistic License 2.0.