Annotate-a-genome: Difference between revisions

From QiuLab
Jump to navigation Jump to search
imported>Weigang
imported>Weigang
Line 71: Line 71:


=Protocol=
=Protocol=
==Dependency==
==Dependencies==
* BASH (default shell of Linux OS and Apple OS X)
* BASH (default shell of Linux OS and Apple OS X)
* Perl and [http://bioperl.org BioPerl]
* Perl and [http://bioperl.org BioPerl]

Revision as of 22:38, 13 February 2014

Project Goals

  • Annotate and add newly sequenced Borrelia genomes to BorreliaBase
  • Build an informatics pipeline for gene prediction, ortholog calls, databasing, and synteny analysis

Download genome sequences from GenBank

Strain Species Group Genome Sequences Notes
B31 B. burgdorferi (reference genome) Lyme Disease Example
CA382 B. burgdorferi (California) Lyme Disease Example
CA8 B. burgdorferi (California) Lyme Disease Example
BgVir B. garinii (Russia) Lyme Disease Example
NMJW1 B. garinii (China) Lyme Disease Example
HLJ01 B. afzelii (China) Lyme Disease Example
Ly B. duttonii (Tanzania) Relapsing Fever Example
A1 B. recurrentis (Ethiopia) Relapsing Fever Example
DAH B. hermsii (Washington State) Relapsing Fever Example
91E135 B. turicatae (Texas) Relapsing Fever Example
Achema B. crocidurae (Mauritania) Relapsing Fever Example
HR1 B. parkeri (??) Relapsing Fever Example
LB-2001 B. miyamotoi (Northeast US) Relapsing Fever Example

Protocol

Dependencies

Fetch genome sequences

  • Commands:
./bioseq -z 'b31_accession' -o 'genbank' > b31.gb # Reference genome for ortholog identification. Choose main, cp26, or lp54
./bioseq -z 'gb_accession' -o 'genbank' > new.gb
./gb2fas -n b31.gb > b31.nuc # Extract CDS
./gb2fas -n new.gb > new.nuc
./bioseq -t b31.nuc > b31.pep # Translate (and remove those with internal stop codons)
./bioseq -t new.nuc > new.pep
  • Perl code for "gb2fas.pl":
#!/usr/bin/env perl
# Extract sequences from a GenBank file
# Input: a GenBank file
# Output: -n: CDS sequences in FASTA; -t: CDS information in Tab-delimited
use strict;
use Bio::SeqIO;
use Getopt::Std;
use Data::Dumper;
use 5.10.0;

my %opts;
getopts('tn',\%opts);

die "$0 [-nt] <genbank_file>\n" unless @ARGV == 1;
my $gb_file = shift @ARGV;
my $in = Bio::SeqIO->new(-file=>$gb_file, -format=>'genbank');
my $cds_ct=0;
while (my $seqobj = $in->next_seq() ) {
    my @features = $seqobj->get_SeqFeatures(); # just top level
    foreach my $feat ( @features ) {
	next unless $feat->primary_tag eq "CDS";
	$cds_ct++;
	&to_db($feat, $cds_ct) if $opts{t};
	&to_nt($feat, $seqobj) if $opts{n};
    }
}

exit;

sub to_nt {
    my $ft = shift;
    my $seq = shift;
    say ">", $ft->get_tag_values("locus_tag");
    my $subseq = $seq->trunc($ft->start, $ft->end);
    if ($ft->strand > 0) {
	say $subseq->seq();
    } else {
	say $subseq->revcom()->seq();
    }
}

sub to_db {
    my $ft = shift;
    my $ct = shift;
    my $orf_id = sprintf "ORF%04d", $ct;
    my $gid = 401; # this is bad and needs improvement: nothing should be hard-coded
    my $con_id = 111114823; # the same problem
    my $locus =  sprintf "%s", $ft->get_tag_values('locus_tag');
    my $prod = sprintf "%s", $ft->get_tag_values('product');
    $prod =~ tr/\'/_/;
    my $strand = ($ft->strand > 0) ? 't' : 'f';
    say join "\t", ($gid, $con_id, $orf_id, 'f', $ft->start, $ft->end, $strand, $locus, $prod);
}

Predict orthologs with reciprocal BLAST

  • Commands:
makeblastdb -in b31.pep -parse_seqids # Prepare the reference DB
makeblastdb -in new.pep -parse_seqids # Prepare the new genome DB
blastp -query new.pep -db b31.pep -outfmt 6 -evalue 1e-3 -out forward_blast.out # Forward BLAST
blastp -query b31.pep -db new.pep -outfmt 6 -evalue 1e-3 -out reverse_blast.out # Reverse BLAST
./check-reciprocal.pl forward_blast.out reverse_blast.out > new.orthlogs 2> new.not-orthologs # Identify orthologs
  • Code for "check-reciprocal.pl":
#!/usr/bin/env perl
use strict;
use warnings;
use Data::Dumper;

die "$0 <forward_blast_output> <reverse_blast_output>\n" unless @ARGV == 2;
my ($fwd, $rev) = @ARGV;
my (@fwd_top_hits, @rev_top_hits);
open FWD, "<" . $fwd;
my (%fwd_top_hits, %rev_top_hits, @query);
my $query_ct=0;
while (<FWD>) {
    chomp;
    my @data = split;
    next if $fwd_top_hits{$data[0]};
    $fwd_top_hits{$data[0]} = $data[1];
    push @query, $data[0];
    $query_ct++;
}
close FWD;
warn "Total query having hits:" . $query_ct . "\n";

open REV, "<" . $rev;
while (<REV>) {
    chomp;
    my @data = split;
    next if $rev_top_hits{$data[0]};
    $rev_top_hits{$data[0]} = $data[1];
}
close REV;

foreach my $q (@query) { # e.g., BafPKo_0002
    my $top = $fwd_top_hits{$q}; # e.g. BB_0002
    if ( $q eq $rev_top_hits{$top}) {
	print "Found reciprocol top hits:\t", $q, "\t", $top, "\n";
    } else {
	warn "Not reciprocol top hits:\t", $q, "\t", $top, "\t", $rev_top_hits{$top}, "\n";
    }
}

exit;

Verify with synteny broswer

./gb2fas -t new.gb > new-to-orf-table.txt