Revision history for GO-TermFinder group of modules.
0.86 Thu Nov 19 09:25:53 2009
Constrain the p-value to not be more than 1, which sometimes
happens due to rounding errors on some platforms.
0.85 Thu Oct 29 12:54:55 2009
Add a missing include file into Distributions.cxx that gets
complained about on some platforms
0.84 Thu Oct 29 11:26:46 2009
Added in some additional error checking, that prevents the C++
code that calculates the probability from being called with
invalid arguments, which had caused some segfaults on some
platforms, and some p-values greater than 1 on others. Not
clear how the C++ code was being called incorrectly, but this
should at least provide a more informative error.
0.83 Thu Jan 22 12:51:03 2009
Changed the way in which genes are rendomly chosen from the
background distribution, when calculating FDR, due to a bias
that existed in the algorithm. Thanks to Peter Wentzell for
pointing this out; also see:
Flight, R.M and Wentzell, P.D. (2009). Potential Bias in
GO::TermFinder. Briefings in Bioinformatics, in press.
In my testing, this improves the FDR when having passed in a
list of genes as a background population from which the test
set were drawn. For some reason, it seems to make less
significant (higher FDR) those values calculated when the
background population of genes is not passed in (and is
assumed to be all genes).
0.82 Wed May 14 13:34:01 2008
No functional changes. Regenerated swig wrappers using
1.3.35, in the hope that that might see some of the failures I
am seeing from CPAN testers. I'm not sure if those errors are
specific to a new version of gcc, or to Perl 5.10.0. If the
last version worked for you without any errors, you don't need
this one.
0.81 Tue May 13 16:06:55 2008
- GO::TermFinder
Made totalNumGenes a public method, which returns the total
number of genes that are in the background set of genes from
which the genes of interest were drawn. Unannotated genes are
included in this count. This allowed me to fix a bug in
analyze.pl, where it didn't always report the correct number
of genes in the background.
- GO::AnnotationProvider::AnnotationParser
Fixed bug in handling a name that was ambiguous when
considered in a case-insensitive fashion, but unambiguous when
considered in a case-sensitive fashion. Thanks to Robert
Flight for the bug report.
- GO::OntologyProvider::OboParser
Added additional logic to detect an error in the obo file if a
node and a parent are in different namespaces, as this
previously produced a fatal error with a cryptic message.
- GO::TermFinder
Prevent crash that could occur when there were no GO nodes
tested - assertion that correction factor had to be >=1 was
incorrect in this case, which could cause the script to die.
Now correctly recognizes this. Thanks to Mark Schroeder at
Princeton for pointing this out.
- GO::View
I had previously commented out a check, that resulted in
postscript images always being generated. Now they are only
generated when requested.
GO-TermFinder-Obo.t
Tweaked test to be regex instead of equality, as different
Perls might add different things to exceptions. This fixes a
failure I was seeing on Solaris.
0.80 Thu Mar 22 18:45:29 2007
- GO::OntologyProvider::OboParser
New module, contributed by Shuai Weng of SGD. This allows you
to use the .obo files instead of the .ontology files. The
.ontology files have been deprecated for a while, and are much
bigger than the .obo version, as they contain redundant
information. The ontology parsing module still ships, but
will be deprecated in future versions. The test suite for the
ontology parser no longer ships with the distribution, though
I still test it locally.
Note, the API for the OboParser is slightly different, in that
you have to provide the aspect of the ontology that you want
(P, C or F), which was not the case when parsing the .ontology
files.
All the example code in the examples/ directory has been
updated to work with the obo parser.
- GO::TermFinder
Updated to deal with the loss of the 'unknown' nodes in the
ontology. With the removal of those nodes, direct annotation
to the aspect node, such as biological_process, is meaningful.
Thus, it now tests for significance for genes annotated
directly to the aspect nodes, ignoring indirect annotations to
those nodes.
Added an additional check, such that if you provide a
background population, but none of the genes in your list of
interest are in that population, it dies with a useful error
message (as opposed to the previously useless one).
Corrected a bug in the Bonferroni correction, where it was too
conservative - not sure what I was thinking.
- GO::TermFinderReport::Text
Can now print out the results in tabular form, instead of the
usual text version, to help with automated analyses. Thanks
to Noah Zimmerman for the code changes.
- Code Quality
I have started to use Perl::Critic locally, and all the code
at least passes level 5. Future revisions will pass harsher
review (with some rules commented out!).
- Compilation
The native/Makefile.PL has been modified to hopefully make it
compile correctly under Cygwin - thanks to Noah Zimmerman for help
with testing.
0.72 Thu Jul 27 16:44:14 2006
- GO::TermFinder
Forgot to have the __saveVariables method save the discarded
genes when simulations were being run, or the FDR was being
calculated. Consequently, this information was being lost if
either FDR or simulations were requested. This is now fixed.
Found a minor buglet, concerning how genes were selected for
simulations if a population had been used. This didn't affect
the actual results, but now is *slightly* faster.
- General
Started using the Test::Spelling module locally, and it found
a few typos that have been corrected.
0.71 Sun Jul 23 16:15:20 2006
- native/Makefile.PL
Fix for the compilation of the C++ code that carries out the
math. It is now hard-coded to use g++ as the LD and CC
variables, and will complain if it can't find g++. If it
fails, it will tell you what needs editing to hopefully fix
it.
- GO::TermFinder
Small amount of refactoring on the findTerms() method, to make
it easier to follow. This doesn't change the functionality.
Added logic such that if a background population is provided,
but then the list of genes for which enriched terms are to be
found has genes not in that background population, then those
genes are discarded from the calculation, and a new method,
discardedGenes, allows the identify of those genes to be
determined.
- tests
Added some new tests, and make sure all code uses strict and
warnings
0.70 Thu Nov 18 11:55:10 2004
- GO::TermFinder
Now uses an entirely new set of code for calculating the
probability based on the hypergeometric distribution (written
by Ihab Awad). The new code is written in C++, and interfaced
to Perl using SWIG. For long running batch jobs, with > 100
lists of genes (using for instance analyze.pl in the examples
directory), the new code is up to 3 times faster. Note, that
installation now requires an ANSI C++ compiler - see the
README for more details.
Note that the ability to use the binomial distribution for the
probability calculation no longer exists - it will always use
the hypergeometric distribution now.
- GO::Utils::File
fixed small bug in that only one leading space would be
stripped from a gene name in GenesFromFile - thanks to Linda
McMahan at Princeton for spotting this one.
- GO::TermFinder
made error message a little more explicit when a goid used to
annotate a gene (as indicated by the annotation provider) does
not appear in the provided ontology - thanks to John Matese
for the suggestion.
- GO::View
small addition that hopefully deals with a problem sometimes
seen when running on Windows, that I think is due to line
endings produced by the dot program.
0.64 Sun Aug 15 23:30:32 2004
- GO::View
Added the ability to create postscript output from
GO::View - simply change the makePs attribute in the
GoView.conf file to a '1', and if you run batchGoView.pl, you
should get a postcript file as well as png or gif images. The
Postscript is not perfect, because the GraphViz interface
doesn't seem to allow me to modify the page size (even though
it says I should be able to), so I have restricted the size of the
image to try and get it to fit on one page. Suggestions as to how to
make the postscript feature more robust would be appreciated.
Added in copious comments to GO::View, so that most of it can
actually be understood my regular programmers - performed some
cleanups or unnecessary or obfuscated code in the process.
Significant refactoring of this module is next on the agenda.
0.63 Wed Aug 11 11:16:23 2004
- GO::View
Fix to bug that was causing some significant nodes to not
have the correct color, and thus making the image somewhat
misleading - thanks to the folks at Princeton for spotting it,
and to Shuai Weng at SGD for providing a fix in record time.
fixed bug in reading of configuration file, that resulted from
minMapWidth being a substring of minMapWidth4OneLineKey. This
now stops a warning being printed.
Added some comments to GO::View, as a start to begin to fully
comment all the code - there's a long way to go until I
understand fully how GO::View works. Then I can add
postscript output for publication quality figures...
0.62 Wed Jul 28 11:46:08 2004
No major new functionality - small bug fix release
- AnnotationParser.pm
fixed capitalization bugs when calling goidsByDatabaseId() in
nameIsAnnotated() instead of goIdsByDatabaseId() on lines 1592
and 1617 - thanks to lfriedl@cs.umass.edu for spotting this.
- GoView.conf
Fixed typo in GoView.conf : totalNumGene should be
totalNumGenes - thanks to John Matese for spotting this one.
- GO::TermFinder
Made __databaseIds() method of GO::TermFinder public, and
named it genesDatabaseIds, as I realized it is the only way
that a client can determine how many genes were actually used
when calculating p-values. Before, I was using the number of
genes that were passed in, but they will get collapsed if more
than one maps to the same databaseId.
- batchGOView.pl
Modified batchGOView.pl to use the genesDatabaseIds method.
Removed its use of the CategorizeGenes function in
GO::Utils::General, as I don't think was was any reasonable
logic for using it (that I can remember).
- GO::TermFinder
Fixed lots of spurious warnings that were due to checking the
state of a variable before it had been set, when using a user
defined background population - thanks to Jeremy Gollub for
spotting this one.
- GO::TermFinder
If a gene is padded in multiple times in either the list of
genes of interest, or the background population, you should
only be warned about that gene once for each list, rather than
every time the gene is encountered.
0.61 Fri May 7 11:23:36 2004
- GO::TermFinder
Made one line optimization to calculation of p-value using the
hypergeometric distribution, such that it is on average about
twice as fast.
0.60 Wed May 5 15:04:06 2004
- GO::TermFinder
The multiple hypothesis correction is no longer done using the
custom method, but instead uses a Bonferroni correction.
Correcting p-values by running simulations is now available,
to allow you to control the Family Wise Error rate.
Added the ability to calculate the False Discovery Rate, as a
potential means of avoiding the whole p-value problem.
see the pod for GO::TermFinder, as well as docs/GO-TermFinder.doc
for more information on these.
- tools
updated the batchGOView.pl tool, and the analyze.pl tool, in
the examples directory, to support printing out of the False
Discovery Rate if calculated. They also both use the newly
created GO::TermFinderReport::Text object, to consolidate code, and
to keep their reports consistent with one another.
- README
Tried to make it a little clearer as to how to install the libraries.
0.50 Tue Dec 16 17:34:50 2003
Big news is that a version of Shuai Weng's (from SGD) GO::View
module has been added in, which can create a graphic
representation of the results of GO::TermFinder. This gives a
much better way to intuitively look at the results. A bunch
of other stuff was added in to support this, including a batch
processor that will generate an html pages with a graphic for
any number of input lists of genes. See batchGOView.pl and
examples.html in the examples directory.
- GO::TermFinder
Fixed a nasty bug that was actually due to a bug in Perl (I
swear!) that meant that clients of the GO::TermFinder would
have a runtime error if they did not recognize one or more of
the genes that it was provided with. This should no longer
happen, and I have written additional tests to make sure it
never happens again!
0.40 Tue Dec 2 18:41:10 2003
- GO::TermFinder
Added in the ability to define a subpopulation of genes as the
background from which the interesting genes were drawn. See
pod for constuctor of GO::TermFinder for more details.
0.30 Wed Nov 26 11:48:57 2003
- GO::AnnotationProvider::AnnotationParser
Extensive reworking of the code, such that it is now case
insensitive, with certain caveats. See the pod for that
module for more details.
- GO::TermFinder
No longer considers the root node, or its child (the aspect)
as hypotheses, as they are known a priori to have a p-value of
1.
- GO::Node
Added lengthOfShortestPathToRoot and meanLengthOfPathsToRoot
methods.
- added in new tests for the AnnotationParser that make sure
that the behaviour with respect to case insensitivity is
correct.
0.23 Sat Nov 1 11:56:43 2003
- GO::OntologyProvider::OntologyParser;
Added in check that a given non-comment line can actually have
a GOID extracted, which makes the error a more informative
when such a line is encountered, due to an error in an
ontology file.
0.22 Sun Oct 19 16:54:38 2003
- GO::TermFinder
Fix for test that could occasionally fail that relied on
the sort order of the pValue array when two items had the
same pvalue. It now sorts such cases explicitly by goid,
so the result should always be the same.
0.21 Thu Oct 16 19:00:14 2003
- GO::TermFinder
Fix for situation when a gene identifier wasn't recognized,
but wasn't handled properly - thanks to Shuai Weng for bring
it to my attention.
Cleaned up the code that calculates the p-values to make it
easier to write a test-suite, which will allow me to make
other desired changes with more confidence.
- tests
Created tests for the GO::TermFinder module itself, that pass
under both OSX and Solaris in my testing - this should make it
much easier to modify in future with less concern for failing
to notice new bugs.
- examples
Added a new example, that makes it easier to analyze multiple
files of gene names.
- suppressed some warnings that occured due to some undefs
being used.
- fixed some documentation typos
0.2 Mon Apr 14 17:55:28 2003
- added in code such that the findTerms() method will return
enough data for you to be able to work out which genes in
the list you provided were annotated to which GO nodes.
- started some work on Annotation, AnnotatedGene, and
Reference objects. Nothing is using them yet though.
- added in over 100 tests(!). Still lots more to do, but this
should help keep the code honest...
0.1 Fri Mar 7 13:51:32 2003
- original version; created by h2xs 1.21 with options
-n GO-TermFinder -X