Many times in bioinformatics we need to deal with huge datasets which  are more than 100GB size. The traditional way to analysis a file is using the while loop

while (FILE){

Do something;

}

This is very slow(since we are using only one processor) and if we have 500 million lines in the dataset it takes more than a day to iterate through the whole dataset. So how do we make best use of all our processors and get the work done quickly?

Here is a very simple and efficient technique with perl which i have been using. I am  more inclined towards using perl fork than perl threads.

One of the oldest way to fork is

my $fork = fork();
if($fork){   
push (@childs,$fork); 
}
elseif($fork==0){
your code here;
exit(0);
}
else{die “Couldnt fork : $!”;}

## wait for the child process to finish
foreach(@childs){
my $tmp=waitid($_,0);
}

what a fork does is it creates a child process and takes the variables and code with it to analyze it separately (detached from the parent process) and thus a separate process is created( which usually runs on a separate processor). Thats it!! One big disadvantage of forking is its very difficult to share variables among the different processes. I will show you how to do it easily but still it has its own drawbacks.

Okie, now if you really do not want to use fork in your code, that’s okie too..There are many useful modules which do it for you very efficiently. One really useful module is Parallel::ForkManager. You can use Parallel::ForkManager to manage the number of forks you want to generate (number of processors you want to use).

Simple usage:
use Parallel::ForkManager;
my $max_processors=8;
my $fork= new Parallel::ForkManager($max_processors);
foreach (@dna) {
$fork->start and next; # do the fork
you code here;
$fork->finish; # do the exit in the child process
}
$pm->wait_all_children;

so you will be generating 8 forks which do the same thing for your each element of array. when one child finishes, Parallel::ForkManager generates a new one and thus you will be using all your processors to analyze the data. Now, if you have generated 8 child processes and want to write the data to one file. You need to lock the file to do this, because you will have problems with the buffering. You can lock the file using flock command.

open (my $QUAL, “myfile.txt”);
flock $QUAL, LOCK_EX or die “cant lock file $!”;
print $QUAL “$output”;
flock $QUAL, LOCK_UN or die “$!”;
close $QUAL;

I would not suggest using flock when dealing with multiple processes because it will decrease the processing efficiency( each child process must wait for the lock to be released by the other child process). Instead, I would suggest each fork writing to a separate file and after the processing just concatenating them.

Putting it all together, If you have 100GB data you can do this

step 1 : split the dataset equally according to number of processors you have. this may take a few hours(about 2-3 hrs for 100GB file)
You can use unix “split” command for this
for example:
my $number_split=int($number_of_entries_in_your_dataset/$max_processors);
my $split_Files=`split -l $number_split “your_file.fasta” “file_name”`;

step2: open you directory comtaining you split files and start Parallel::ForkManager.
For example:
opendir(DIRECTORY, $split_files_directory) or die $!; ### open the directory
my $fork= new Parallel::ForkManager($max_processors);
while (my $file = readdir(DIRECTORY)) { ### read the directory
if($file=~/^\./){next;}
print $file,”\n”;
########## Start fork ##########
my $pid= $super_fork->start and next;
Whatever you want to do with the split file ;
analyze my piece of $file;
######### end fork ###############
$super_fork->finish;
}
$super_fork->wait_all_children;

So basically each processor will be active with its piece of data (split file) and thus you have created 8 processes at one time which run without interfering with the other process. I again will not suggest writing output from each child process to one file(for reasons above). Write output from each fork to a separate file and finally concatenate them. Thats it, you have just increased your program speed by 8 times!! Isnt it easy?

Note:
You may worry about concatenation of the output each child generates, since it does take some time(remember 100GB). I think now you can use a mysql database LOAD DATA LOCAL INFILE command to load all the files into a single table(Should take about 3hrs for 100Gb dataset) and then export the whole table into one file. This should be faster than just concatenating them using “cat” command.(correct me if I am wrong)

Or much simpler way is to use pipes

cat output_dir/* | my_pipe or my_pipe <(file1) final_file;

Thats it guys!! Enjoy programming and please do comment. I am not a computer scientist so forgive me for any mistakes and if any please report them. Thank you.

Special Symbols – File handlers

$@ Perl error string

$! Error number from C, ‘errno’

$^E Extended OS error info, such as ‘CDROM tray not closed’

$? Exit status from last process

$AGRV – name of current file

@ARGV – command line arguments

$ARGV – special file handle for command line filenames

$. – current line number

$/ - input line delimiter

$\ - output line delimiter

$% - current page number

$&/${^MATCH} – last successful matching string

$`/${^PREMATCH} – the string preceding the last matching string

$’/${^POSTMATCH} – the string following the last matching string

$1, $2, … - matching groups in the parentheses in pattern

More at https://www.tutorialspoint.com/perl/perl_special_variables.htm

To resolve this, download and have in SSPACED-Longreads folder.

Cheers :)

Canu is a hierachical assembly pipeline which runs in four steps:

• Detect overlaps in high-noise sequences using MHAP
• Generate corrected sequence consensus
• Trim corrected sequences
• Assemble trimmed corrected sequences

Read the documentation

New release https://github.com/marbl/canu/releases

Address of the bookmark: https://github.com/marbl/canu

ORFfinder is a graphical analysis tool for finding open reading frames (ORFs). We’ve been working on a few updates, and we’d like to find out what you think about them. Read on to find out what you can do with the new ORFfinder.

Smart BLAST (https://ncbiinsights.ncbi.nlm.nih.gov/2015/07/29/smartblast/)

Select one or a group of ORFs and BLAST several databases at once, and use the newly developed SmartBLAST to verify protein names. Looking for the traditional results from BLAST? They’re there too.

Mauve has been developed with the idea that a multiple genome aligner should require only modest computational resources. It employs algorithmic techniques that scale well in the lengths of sequences being aligned. For example, a pair of Y. pestis genomes can be aligned in under a minute, while a group of 9 divergent Enterobacterial genomes can be aligned in a few hours. However, the current algorithm’s compute time (progressiveMauve) scales cubically in the number of genomes to align, making it unsuitable for datasets containing more than 50-100 bacterial genomes.

Address of the bookmark: http://darlinglab.org/mauve/mauve.html

Tool at https://github.com/kbseah/genome-bin-tools

Address of the bookmark: http://journal.frontiersin.org/article/10.3389/fmicb.2015.01451/full