I'd like to find transcription start sites for Ensembl genes (mouse, build 67, NCBIM37). Are these annotations available from Ensembl's site, and if so, what file(s) should I be looking at?

Via FTP

Here's the script I wrote to pull mouse exons from Ensembl, release 67 via FTP and filter them for TSS positions. I treat the start position of the first forward strand exon and the stop position of the first reverse strand exon as the position of the TSS:

#!/usr/bin/env perl

use strict;
use warnings;
use File::Path;

sub trimWhitespace($)
{
    my $string = shift;
    $string =~ s/^\s+//;
    $string =~ s/\s+$//;
    return $string;
}

sub trimQuotes($)
{
    my $string = shift;
    $string =~ s/^"//;
    $string =~ s/"$//;
    return $string;
}

my $ensemblURL = "ftp://ftp.ensembl.org//pub/release-67/gtf/mus_musculus/Mus_musculus.NCBIM37.67.gtf.gz";
my $dataDir = "../data"; if (! -d $dataDir} { mkpath $dataDir; }
my $ensemblGzFn = "$dataDir/Mus_musculus.NCBIM37.67.gtf.gz";
my $ensemblBedFn = "$dataDir/Mus_musculus.NCBIM37.67.gtf.bed";

if (! -e $ensemblGzFn) {
    print STDERR "retrieving mouse records...\n";
    system ("wget --output-document=$ensemblGzFn $ensemblURL") == 0 or die "could not retrieve Ensembl records from URL: $ensemblURL\n";
}

if (! -e $ensemblBedFn) {
    print STDERR "converting mouse records from GTF to sorted BED...\n";
    system ("gunzip --stdout $ensemblGzFn | gtf2bed - | sort-bed - > $ensemblBedFn") == 0 or die "could not convert Ensembl record to sorted BED\n";

    open my $allRecFh, "<", $ensemblBedFn or die "could not open all records\n";
    while (<$allRecFh>) {
        chomp;
        my ($chr, $start, $stop, $id, $score, $strand, $source, $feature, $frame, $attributesStr) = split("\t", $_);

        if ($feature eq "exon") {
            my ($tssChr, $tssStart, $tssStop, $tssId, $tssScore, $tssStrand);
            my @attributes = map (trimWhitespace($_), split(";", $attributesStr));
            my $attributesRef;
            foreach my $attributePair (@attributes) {
                my @pairItems = split(" ", $attributePair);
                my $key = $pairItems[0];
                my $value = trimQuotes($pairItems[1]);
                $attributesRef->{$key} = $value;
            }
            # find TSS start position
            if (exists $attributesRef->{"gene_name"}) {
                $tssChr = "chr$chr";
                if ($strand eq "+") {
                    $tssStart = $start;
                    $tssStop = $tssStart + 1;
                }
                elsif ($strand eq "-") {
                    $tssStart = $stop;
                    $tssStop = $tssStart + 1;
                }
                $tssId = $attributesRef->{"gene_name"};
                $tssScore = $attributesRef->{"exon_number"};
                $tssStrand = $strand;

                my $tssGeneId = $attributesRef->{"gene_id"};

                if ($tssScore == 1) {
                    print STDOUT join("\t", ($tssChr, 
                                             $tssStart, 
                                             $tssStop,
                                             $tssId,
                                             $tssScore,
                                             $tssStrand,
                                             $tssGeneId
                                             ))."\n";
                }
            }
        }
    }
    close $allRecFh;
}

In running this FTP-sourced data search, I get 97639 records (which matches Malachi's answer via BioMart):

$ ./get_mm9_ensembl_tss_records.pl > ../data/mm9.ensembl67.tss.bed
$ wc -l ../data/mm9.ensembl67.tss.bed
97639

Records seem to match up:

$ grep Gm16627 ../data/mm9.ensembl67.tss.bed
chr17   71274336        71274337        Gm16627 1       -       ENSMUSG00000085299

$ grep Vmn2r-ps113 ../data/mm9.ensembl67.tss.bed
chr17   18186447        18186448        Vmn2r-ps113     1       +       ENSMUSG00000067978
chr17   18201000        18201001        Vmn2r-ps113     1       +       ENSMUSG00000067978

I'll mark both answers as correct, since Malachi gives me a "sanity check" result and Emily provided the FTP URL to the Ensembl mouse data.

Via Perl API

From the Ensembl-dev mailing list, I was pointed to the following version of the ensembl portion of the Ensembl Perl API setup process:

http://www.ensembl.org/cvsdownloads/ensembl-67.tar.gz

The remaining components were installed per Emily's instructions in her answer below.

Additionally, the useastdb.ensembl.org host only supports Releases 71 and 70. It is necessary to use ensembldb.ensembl.org to access older releases or host the databases locally.

After making these changes to my setup, here is the script I used to pull all exons for all genes from mouse release 67, printing them in a BED6-like format:

#!/usr/bin/env perl

use strict;
use warnings;
use Data::Dumper;
use Bio::EnsEMBL::DBSQL::DBAdaptor;

my $host   = 'ensembldb.ensembl.org';
my $user   = 'anonymous';
my $dbname = 'mus_musculus_core_67_37';
my $port   = 5306;

my $db = new Bio::EnsEMBL::DBSQL::DBAdaptor(-host => $host,
                                            -user => $user,
                                            -dbname => $dbname,
                                            -port => $port);

my $slice_adaptor = $db->get_SliceAdaptor();

my $slices = $slice_adaptor->fetch_all('chromosome');
foreach my $slice (@{$slices}) {
    my $chr = "chr".$slice->seq_region_name();
    my $genes = $slice->get_all_Genes();
    foreach my $gene (@{$genes}) {
        my $exons = $gene->get_all_Exons();
        my $id = $gene->external_name();
        my $exon_index = 1;
        my $exon_number = $exon_index;
        my $exon_count = scalar(@{$exons});        
        foreach my $exon (@{$exons}) {
            my $start = $exon->start();
            my $end = $exon->end();
            my $stable_id = $exon->stable_id();
            my $strand = $exon->strand();
            if ($strand == 1) { 
                $strand = "+";
                $exon_number = $exon_index;
            } 
            elsif ($strand == -1) { 
                $strand = "-";
                $exon_number = $exon_count - $exon_index + 1;
            } 
            else { 
                die "unknown value for strand\n"; 
            }
            print STDOUT join("\t", ($chr, $start, $end, $id, $exon_number, $strand))."\n";
            $exon_index++;
        }
    }
}

This does not fully recapitulate the GTF output, and access to this remote database is slow, but it should be enough to filter the output from this script on the score column, retrieving only 1-scored elements and looking at the strand in order to get the position of the TSS, using awk or something else.

I'll need to verify that this can output the correct number of single-exons-via-genes, and locally hosting databases is probably a good idea in order to get a quick answer, but this is probably a pretty decent start.