Difference between revisions of "Talk:Perl"

From Christoph's Personal Wiki
Jump to: navigation, search
 
(3 intermediate revisions by the same user not shown)
Line 1: Line 1:
==Perl and [[MySQL]]==
 
This section will just list a bunch of random examples. They are useful to give you an idea of what you can do with Perl, MySQL, and the [[:Category:Linux Command Line Tools|CLI]].
 
  
''Note: These examples were taken from my course in Comparative Microbial Genomics at CBS (in Denmark).''
 
 
<pre>
 
mysql -B -e "update cmp_genomics.features set note = '' where user = USER() and note not like 'tcs%' or note is null"
 
foreach accession (AE017042 AE016879 AL111168 AL645882 AP008232 AP009048 BA000021 CP000034)
 
  foreach type (ecf s54 s70)
 
    cat source/$accession.proteins.$type.sigmas.hmmsearch | \
 
    perl -ne 'next unless /^CDS_(\d+)\-(\d+)_DIR([\-\+]+)\s+([0-9\-\.e]+)\s+([0-9\-\.e]+)\s+/;\
 
              my ($start,$stop,$dir,$score,$evalue) = ($1,$2,$3,$4,$5);\
 
              next unless $score > 0;\
 
              print "update cmp_genomics.features set note = \"Sigma Factor '$type'\" \
 
                      where start=$start and stop = $stop and user = user() and accession = \"'$accession'\";\n";'\
 
    | mysql
 
  end
 
end
 
</pre>
 
<pre>
 
mysql -B -e "update cmp_genomics.features set note = '' where user = USER() and note not like 'sigma%' or note is null"
 
foreach accession (AE017042 AE016879 AL111168 AL645882 AP008232 AP009048 BA000021 CP000034)
 
  foreach type (RRreciever HisKA_1 HisKA_2 HisKA_3 HWE_HK)
 
    cat source/$accession.$type.TCS.hmmsearch | \
 
    perl -ne 'next unless /^CDS_(\d+)\-(\d+)_DIR([\-\+]+)\s+([0-9\-\.e]+)\s+([0-9\-\.e]+)\s+/;\
 
              my ($start,$stop,$dir,$score,$evalue) = ($1,$2,$3,$4,$5);\
 
              next unless $score > 0; \
 
              print "update cmp_genomics.features set note = \"TCS '$type'\" \
 
                      where start=$start and stop = $stop and user = user() and accession = \"'$accession'\";\n";'\
 
    | mysql -B
 
  end
 
end
 
</pre>
 
<pre>
 
rm -f source/all.16s.fsa
 
foreach accession (AE017042 AE016879 AL111168 AL645882 AP008232 AP009048 BA000021 CP000034)
 
  mysql -B -N -D cmp_genomics -e "select substr(seq,start,stop-start+1) as sequence,\
 
    replace(substring_index(genomes.organism,' ',4),' ','_') as name,dir from genomes,features where \
 
    genomes.accession = features.accession and genomes.user = features.user and \
 
    features.user = user() and featuretype='rRNA' and features.accession = '$accession'" | \
 
  perl -ane '$F[0] = reverse ( $F[0] ) if $F[2] eq "-";$F[0] =~ tr/ATGC/TACG/ if $F[2] eq "-";\
 
              print ">$F[1]\n"; \
 
              for ( $x = 0 ; $x < length ( $F[0] ) ; $x = $x+60) { print substr ( $F[0] , $x , 60),"\n"; } last;' \
 
  >> source/all.16s.fsa
 
end
 
less source/all.16s.fsa
 
</pre>
 
<pre>
 
cat ~/Ex3/source/CP000034.codon-usage.aa.list \
 
| perl -ane 'next unless $F[1] =~ /^[A-Z]+/;print "\t$F[1]"; ' \
 
> aa.dat
 
mysql -B -N -e 'select accession,organism from cmp_genomics.genomes where user=USER()' \
 
> organism.names
 
foreach accession (AE017042 AE016879 AL111168 AL645882 AP008232 AP009048 BA000021 CP000034)
 
  cat ~/Ex3/source/$accession.codon-usage.aa.list \
 
  | perl -ane 'BEGIN { foreach (`cat organism.names`) \
 
                      {chomp; @A = split (/\t+/,$_);$NAMES{$A[0]} = $A[1];} } \
 
              next unless $F[1] =~ /^[A-Z]+/;$FREQ{$F[1]} = $F[3]; push @AA,$F[1]; \
 
              END { print "\n"; print "$NAMES{'$accession'}"; foreach $aa (@AA){print "\t",$FREQ{$aa};}}' \
 
  >> aa.dat
 
end
 
</pre>
 
<pre>
 
mysql -e "delete from localization where user=USER() and type='signalp'" cmp_genomics
 
foreach accession (AE017042 AE016879 AL111168 AL645882 AP008232 AP009048 BA000021 CP000034)
 
  perl -ne 'if(/^CDS_(\d+)-(\d+)_DIR([+-])\s+0.\d+\s+\d+\s+[YN]\s+0.\d+\s+\d+\s+[YN]\s+[10].\d+\s+\d+\s+Y\s+[10].\d+\s+Y\s+[10].\s+\s+Y\s+CDS_\d+-\d+_DIR[+-]\s+Q\s+[10].\d+\s+[10]\s+Y\s+[10].\d+\s+Y/) \
 
            {print "insert into localization (user, protid, type) \
 
                          select USER() as user, ft.id as protid, \"signalp\" as type \
 
                                  from features as ft \
 
                                  where ft.user=USER() and \
 
                                        ft.accession = \"'$accession'\" and \
 
                                        ft.start = $1 and ft.stop = $2 and \
 
                                        ft.dir = \"$3\";\n";}' $accession.signalp \
 
  | mysql -D cmp_genomics
 
end
 
</pre>
 
<pre>
 
perl -e 'foreach $i (1..11) { open IN , "col$i"; while (<IN>) { chomp;($a , $c) = split (/\t+/,$_); $DATA{$a}{$i} = $c;} close IN;} \
 
        END {foreach $a ( reverse sort keys %DATA) {foreach $i (1..11) {$DATA{$a}{$i} = 0 if $DATA{$a}{$i} eq "" and $i > 1; \
 
        print $DATA{$a}{$i}; print "\t" if $i < 11;}print "\n";}}' \
 
> final.dat
 
 
R
 
postscript ('final.ps')
 
source ( "/home/people/kiil/Exercises/files/heatmap.R" )
 
data <- as.matrix ( read.table ( 'final.dat' , row.names=1, header= TRUE,sep = "\t" ) )
 
pal <- maPalette(low = rgb(10,10,10,maxColorValue = 10), high = rgb(0,0,10,maxColorValue = 10), mid= rgb(5,5,10,maxColorValue = 10),, k = 1024)
 
par(oma=c(0, 0, 0, 0.5))
 
data  <- scale(data)
 
new_heatmap ( data , cexCol=1,cexRow=1,colors = pal)
 
dev.off()
 
quit(save="no")
 
</pre>
 
 
==Perl and bioinformatics==
 
*Complement every base in the file foo.fsa:
 
perl -pe 'tr/ATCG/TAGC/' foo.fsa
 
*Add first and fourth columns:
 
perl -ane 'print $F[0] + $F[3], "\n"' datafile
 
*Only print lines 15 - 17:
 
perl -ne 'print if 15 .. 17' *.gff
 
*Count lines:
 
perl -ne '$counter++; END { print "$counter lines"; }' datafile
 
*Split a multi-sequence FASTA file into individual files:
 
perl -ne 'BEGIN{ $/=">"; } if(/^\s*(\S+)/){ open(F,">$1.fsa")||warn"$1 write failed:$!\n";chomp;print F ">", $_ }'
 
 
==Transpose matrix==
 
''Note: The following could be used to transpose a CSV file.''
 
<pre>
 
#!/bin/sh
 
# Example CSV file
 
# FROM:
 
#  a1,a2,a3
 
#  b1,b2,b3
 
# TO:
 
#  a1,b1
 
#  a2,b2
 
#  a3,b3
 
 
perl -e '$unequal=0;$_=<>;s/\r?\n//;@out_rows=split(/\,/, $_);\
 
$num_out_rows=$#out_rows+1;while(<>){s/\r?\n//;@F=split(/\,/,$_);\
 
foreach $i (0 .. $#F){$out_rows[$i] .="\,$F[$i]";}\
 
if($num_out_rows !=$#F+1){$unequal=1;}} END\
 
{foreach $row (@out_rows){print"$row\n"} \
 
warn "\nWARNING! Rows in input had different numbers of columns\n" \
 
if $unequal;warn "\nTransposed table: result has $. columns and $num_out_rows rows\n\n" }' $1
 
</pre>
 
 
==Misc==
 
*Change columns in each line of files:
 
perl -e '$sep=",";while(<>){s/\Q$sep\E/\t/g;print $_;}' file.csv >file.tab
 
 
*Change tab-separated data to use a different delimiter:
 
perl -e '$sep=",";while(<>){s/\t/$sep/g;print $_;}' file.tab >file.csv
 
 
*Remove spaces in a line:
 
perl -e 'while(<>){s/ //g;print $_;}' foo
 
 
*Change all characters to upper case:
 
perl -e 'while(<>){print uc($_);}' foo
 
 
*Change a FASTA file into tabular format:
 
perl -e '$count=0; $len=0; while(<>){s/\r?\n//;s/\t/ /g;if(s/^>//){if($. !=1){print "\n"}s/ |$/\t/;$count++;$_ .="\t";}else{s/ //g;$len +=length($_)}print $_;}print "\n";' foo.fsa >foo.tab
 
 
*Change a tabular file into FASTA format:
 
perl -e '$len=0;while(<>){s/\r?\n//;@F=split /\t/,$_;print ">$F[0]";if(length($F[1])){print" $F[1]"}print "\n";$s=$F[2];$len+=length($s);$s=~s/.{60}(?=.)/$&\n/g;print "$s\n";}' foo.tab >foo.fsa
 
 
*Change from one biological format to another:
 
perl -MBio::SeqIO -e '$informat="genbank";$outformat="fasta";$count=0;for $infile (@ARGV){$in=Bio::SeqIO->newFh(-file =>$infile, -format =>$informat);$out=Bio::SeqIO->newFh(-format =>$outformat);while(<$in>){print $out $_;$count++;}}' foo.gbk >foo.fsa
 
 
==External links==
 
*[http://sysbio.harvard.edu/csb/resources/computational/scriptome/UNIX/Tools/Change.html]
 

Latest revision as of 08:46, 8 September 2007