sub consensus {
   # Picks an arbitrary result from
   # among the highest in case of tie.

   my %counts;
   $counts{$_}++ for @_;
   return ( sort { $counts{$b} <=> $counts{$a} } keys(%counts) )[0];
}

{
   chomp( my $seq  = <> );
   chomp( my @lbls = <> );

   s/^SEQ:// for $seq;
   s/^LBL:// for @lbls;

   for my $i (0..length($seq)-1) {
      printf("%s %s\n",
         substr($seq, $i, 1),
         consensus( map substr($_, $i, 1), @lbls ),
      );
   }
}
[download]

Alternate output:

   my $consensus = '';

   for my $i (0..length($seq)-1) {
      $consensus .= consensus( map substr($_, $i, 1), @lbls );
   }

   print("SEQ: $seq\n");
   print("LBL: $consensus\n");
[download]

Contrary to rovf and AZed, it's my understanding that the consensus for the "K" at index 1 isn't affected by the labels for the "K" at index 2. Let me know if that's wrong.