NAME
GH::Sim4 - a perl XS encapsulation of the Sim4 alignment tool:
SYNOPSIS
#
use GH::Sim4 qw/ sim4 /;
$genomic = "acgtacgtacgtacgtacgtacgtacgtacgtacgtacgtacgtacgtacgtac";
$cDNA = "acgtacgtacgt";
$result = sim4($genomic, $cDNA);
$result = sim4($genomic, $cDNA,
{"W" => 15, "R" => 1, "RaiseError" => 1});DESCRIPTION
GH::Sim4 is a module that provides direct access to the sim4 cDNA-genomic sequence alignment tool. Sim4 is described in more detail at:
Florea L, Hartzell G, Zhang Z, Rubin GM, Miller W. A computer program for aligning a cDNA sequence with a genomic DNA sequence. Genome Res. 1998 Sep;8(9):967-74.
and
http://bio.cse.psu.edu/
This module has two basic goals: provide the ability to run sim4 from Perl programs without the overhead of starting a separate process and provide access to sim4's results without having to catch and parse its textual output. It also has the added benefit of providing more pleasant (for a Perl-ite at least) error handling than is easily achievable by firing off the command line version.
EXPORT
Nothing is EXPORTed by default.
EXPORT_OK
sim4
ARGUMENTS
sim4() has two required arguments and a third optional argument. The first argument is a scalar variable containing the genomic sequence. The second argument is a scalar variable containing the cDNA sequence. The third (optional) argument is a hash containing option settings.
Option settings
Sim4 options are set by including their name and value in the optional (third) argument to the function call. See the SYNOPSIS above for an example. Most of the options have the same names, meanings and constraints as they have in the stand-alone version of sim4.
Here is their description, cribbed from sim4's usage message. Default values are in parentheses:
W - word size. (W=12)
X - value for terminating word extensions. (X=12)
K - MSP score threshold for the first pass. (e.g., K=16)
C - MSP score threshold for the second pass. (e.g., C=12)
R - Search the direct sequence or reverse complement of the cDNA: 0 - search the '+' (direct) strand only; 1 - search the '-' strand only; 2 - search both strands and report the best match.
The coordinates of the matches that sim4 reports are in the cDNA sequence that gave the better alignment. In other words, for a forward match, position 10 is the 10th base in the cDNA sequence as it was handed to sim4(). For a reverse match, position 10 is the 10th base in the reverse complement of the cDNA sequence, which corresponds to position "length of the cDNA minus 10 minus 1" in the original sequence. (R=2)
D - bound for the range of diagonals within consecutive msps in an exon. (D=10)
H - weight factor for MSP scores in relinking. (H=500)
P - if not 0, remove poly-A tails; report coordinates in the '+' (direct) strand for complement matches; use lav alignment headers in all display options. (P=0) [XXXX BUG? Check out how coord's are returned!] Poly-A tails are recognized *before* the alignment occurs and ignored during the alignment process.
N - accuracy of sequences (non-zero for highly accurate). (N=0)
B - if 0, dis-allow ambiguity codes (other than N and X) in the sequence data. (B=1)
A - must be 0 or 1. If A=1, sim4()) will include a textual representation (like that generated by sim4 with its A=1 setting) in the hash of results under the key "alignment_string".
The S option is not supported. Setting it will result in the call to sim4() failing (returning NULL, generating an exception, and/or printing a warning, depending on the how error handling is configured).
Two additional options control how errors are handled (modeled on the Perl DBI interface):
PrintError Setting "PrintError" to a true value will cause the module to make a warn() call if an error occurs. By default this will cause the error message to be printed to the standard error stream, but it can be caught using the $SIG{__WARN__} handler.
RaiseError Setting "RaiseError" to a true value will cause the module to make a die() call fi an error occurs. By default this will cause the application to exit, but it may be caught inside an eval or via the $SIG{__DIE} handler.
RETURN VALUE
ICK. Given that sim4 succeeds, it's currently returning a big ole' hash, with just about everything that the sim4 algorithm discovers tucked in there somewhere. Not all of the values are guaranteed to be meaningful (or even valid, I'm not always sure what sim4's thinking), but the fundamental information is easy to recognize. When in doubt, see if it agrees with something that the command line sim4 says. The perl Data::Dumper module can be very useful here. Sorry.... Suggestions for particularly useful interfaces would be appreciated.
That said, here are some of the interesting bits:
Exons
The exons element in the hash contains a reference to a list of hashes, one per each exon that sim4 identifies:
'exons' => [
{
'nmatches' => 120,
'flag' => 0,
'length' => 120,
'min_diag' => 0,
'to1' => 170,
'ori' => ' <-',
'to2' => 120,
'max_diag' => 0,
'ematches' => 0,
'alen' => 120,
'from1' => 51,
'edist' => 0,
'from2' => 1,
'match' => 100
},
...
{
'nmatches' => 156,
'flag' => 0,
'length' => 156,
'min_diag' => 0,
'to1' => 1546,
'ori' => undef,
'to2' => 684,
'max_diag' => 0,
'ematches' => 0,
'alen' => 156,
'from1' => 1391,
'edist' => 0,
'from2' => 529,
'match' => 100
}
],Exon orientation (ori).
There are four possible values for the "ori" field in the hash of information about each exon.
-> a.k.a. forward.-
Sim4 reports this orientation when it recognizes the splice sites "GT---AG" in the genomic sequence a the edges of the intron.
<-a.k.a. reverse-
Sim4 reports this orientation when it recognizes the splice sites "CT---AC" in the genomic sequence a the edges of the intron.
--a.k.a. indeterminate-
Sim4 reports this orientation when the splice sites seem equally likely to be either forward or reverse (e.g. the forward score is equal to the reverse score).
==a.k.a. problematic-
Sim4 reports this orientation when this exon and the next one are not contiguous in the cDNA (e.g., if e1 and 2 are neighboring exons, [e2->from2 - e1->to2 - 1 != 0]).
match_orientation
match_orientation is set to either 'forward' or 'reverse', specifying the orientation in which the best match was found (using sim4's standard semantics.
'match_orientation' => 'forward'exon_count
exon_count contains the number of exons that sim4 discovered.
'exon_count' => 7number_matches
number_matches specifies the number of bases that sim4 matched between the cDNA and the genomic sequence.
'number_matches' => 682,coverage_int
coverage_int specifies the number of bases in the genomic sequence (I think...) that were covered by the alignment.
'coverage_int' => 684,coverage_float
coverage_float specifies the ratio of the number of bases covered to the number of bases in the genomic sequence.
'coverage_float' => '1',edit_distance
edit_distance counts the number of edit operations in the alignment.
'edit_distance' => 2,alignment_string
'alignment_string' =>
0 . : . : . : . : . :
51 ATGGTGGGTGTGTCGCCGAAGATCGCTCCGTCGATGTTGTCATCGGACTT
||||||||||||||||||||||||||||||||||||||||||||||||||
1 ATGGTGGGTGTGTCGCCGAAGATCGCTCCGTCGATGTTGTCATCGGACTT
...
Only included in the output when the options include "A" => 1.
Contains the textual representation of the alignment that the stand-alone
sim4 application prints out when its A option is set to 1.exon_alignment_strings
An reference to an array of the textual representations of each exon's
alignment.
Only included in the output when the options include "A" => 1.Opaque/Problematic fields.
Several of the field in the hash are useful, but require that one understand what sim4's doing internally. The alignment operations are a good example. Judicious use of Data::Dumper and perusing the sim4 code (particularly sim4.init.c) should give you a handle on them.
MODULE INTERNALS
The stand alone version of sim4 starts with a main() routine that lives in the file sim4.init.c. It is responsible for loading sequences, processing arguments, and providing results. All of the real work is done by a routine called SIM4(), which lives in the file named sim4b1.c. When it runs into trouble, it calls various flavors of fatal routines that throw up their hands, print errors, and quits.
The perl xs module started life as a minimal wrapper around the SIM4() call and has sprouted many of the features implemented in sim4.init.c (frequently by virtual cut-and-paste). Where something's broken, it's almost certainly my fault.
There are three levels to my implementation. The Perl-ish parts live in Sim4.pm. There's a minimal Perl XS layer, in Sim4.xs, it's kept thin because I find miss the support that the one true editor gives me when I'm working in real Perl or real C. The bottom layer lives in sim4_helpers.c and handles calling into the original SIM4 code and packaging up the results.
Debugging
Tracking down problems in the code can be a bit difficult, since it's dynamically loaded into the perl executable at run time. Here's a methodology that gives you Perl's debugger to poke around the Perl code and gdb for poking at the C code.
Before you begin, make sure that everythings compiled with debugging support.
Makefile.PLneeds to have something like this:'OPTIMIZE' => '-g',Now, set up a simple test case using the standard perl test harness, eg:
use Test; use Data::Dumper; my $result; my $cDNA; my $genomic; BEGIN { plan tests => 3 }; # or how ever many you expect use GH::Sim4 qw/ sim4 /; print "# check that the library loads correctly.\n"; ok(1); # If we made it this far, we're ok. ######################### # # first pass test, just see if it works # print "#################\n# Basic functionality test.\n#\n"; $cDNA = slurp("t/cDNA-1.fasta"); $genomic = slurp("t/genomic-1.fasta"); undef $result; $result = sim4($genomic, $cDNA, {"R" => 0}); print "# check if sim4 returned a defined value.\n"; ok(defined($result)); print "# check that it returned the right number of exons.\n"; ok(scalar @{$result->{exons}}, 7); print "# check that the alignment is on the forward strand.\n"; ok($result->{match_orientation}, 'forward');Next, create a gdb init file (
.gdbinit) (this will save you a lot of typing). This example defines two aliases,randr2that run different scripts (as described above). The particular-Iinclude flags here are correct for my particular system at the moment, but might not be for you. Just crib them from the one Perl uses when you run amake teston the module. Be careful that the run line is either all on a single line or that there's a \ at the end of the line to continue it onto the next.# define r run -Iblib/arch -Iblib/lib -I/usr/lib/perl5/5.6.0/i386-linux \ -I/usr/lib/perl5/5.6.0 -d t/sim4.t end define r2 run -Iblib/arch -Iblib/lib -I/usr/lib/perl5/5.6.0/i386-linux \ -I/usr/lib/perl5/5.6.0 -d /tmp/moose.t endNext, start up perl inside the debugger. Assuming that everythings covered by your path variable, you can just say:
gdb perlNow, start your script. Since the
-dflag is included in the aliases that you've defined in your.gdbinitfile, you'll find yourself sitting at the perl debugger prompt.For example, just use the gdb alias that you defined above:
rStep through the test script (e.g. using
n) until you've passed theuse GH::Sim4line. Everything is now loaded into memory. At this point, you can use a control-C to interupt the perl interpreter and drop into gdb. From there you can set breakpoints, examine variables, continue execution, etc.... The routinesim4_helperis a good entry point into the C interface code, andSIM4is where all of the real sim4 magic begins.At this point, you can move back and forth between the Perl debugger and gdb, stepping and examining to your hearts content.
BUGS
Almost certainly.
For example:
- Inefficiency The command line version of sim4 does some fancy footwork so that various internal data structures are built using the shorter of the two sequences. This can really improve performance. This module doesn't (for the moment) implement that optimization. It will be Real Soon Now.
- Memory leaks. The stand alone implementation version of sim4 handles fatal exceptions by throwing up its hands and quitting, without necessarily cleaning up any resources that it may have allocated. Since the program just exits, these resources are magically freed and life goes on happily. The GH::Sim4 module handles these programmatic hissy-fits by converting them into perl die() and/or warn() calls, but it doesn't do any better at cleaning up the things that sim4's allocated. Once control returns to the module's caller, the references to this information are lost and they are inaccessible. A long-running program that encounters many exceptional situations may suffer from this leakage. This isn't likely to be fixed.
- Lower case sequence characters. The sim4 internals expect the seqeunces to be upper case. There is a routine [seq_toupper()] which will uppercase the sequences for you, but it only checks the first 100 bases of the sequence to decide whether or not to waste the time converting it. This "feature" will remain so that the Perl XS routines are consistent with the command line application.
AUTHOR
George Hartzell, hartzell@fruitfly.org