Kraken taxonomic sequence classification system
===============================================

SPECIAL NOTE TO THOSE UPGRADING FROM v0.9.1b AND EARLIER:

The database structure for Kraken has changed to increase classification
speed.  Versions of Kraken prior to the v1.0 release will still run
against databases created with older versions of Kraken, but the new
databases cannot be run with older versions of Kraken.  See the section
UPGRADING DATABASES TO v0.10+ for more details.


INTRODUCTION
============

Kraken is a taxonomic sequence classifier that assigns a taxonomic
label to short DNA reads.  It does this by examining the k-mers
within a read and querying a database with those k-mers.  This database
contains a mapping of every k-mer in Kraken's genomic library to the
lowest common ancestor (LCA) in a taxonomic tree of all genomes that
contain that k-mer.  The set of LCA taxa that correspond to the k-mers
in a read are then analyzed to create a single taxonomic label for the
read; this label can be any of the nodes in the taxonomic tree.

Kraken is designed to be rapid, sensitive, and highly precise.  Our
tests on various real and simulated data have shown Kraken to have
sensitivity slightly lower than Megablast with precision being slightly
higher.  On a set of simulated 100 bp reads, Kraken processed over 1.3
million reads per minute on a single core in normal operation, and over
4.1 million reads per minute in quick operation.

The latest version of Kraken will be available at the Kraken website:

  http://ccb.jhu.edu/software/kraken/

If you use Kraken in your research, please cite our paper; the citation
is available on the Kraken website.


SYSTEM REQUIREMENTS
===================

Note: Users concerned about the disk or memory requirements should
  read the paragraph about MiniKraken, below.

Disk space: Kraken's standard database will require at least 150 GB
  of disk space.  Customized databases may require more or less space.
  Disk space used is linearly proportional to the number of distinct
  k-mers; as of Dec. 2013, Kraken's default database contains just
  over 5.8 billion (5.8e9) distinct k-mers.

  In addition, the disk used to store the database should be
  locally-attached storage.  Storing the database on a network
  filesystem (NFS) partition can cause Kraken's operation to be
  very slow, or to be stopped completely.  As NFS accesses are 
  much slower than local disk accesses, both preloading and database
  building will be slowed by use of NFS.

Memory: To run efficiently, Kraken requires enough free memory to
  hold the database in RAM.  While this can be accomplished using a
  ramdisk, Kraken supplies a utility for loading the database into
  RAM via the OS cache.  The default database size is 74 GB (as of
  Dec. 2013), and so you will need at least that much if you want
  to build or run with the default database.

Dependencies: Kraken currently makes extensive use of Linux utilities
  such as sed, find, and wget.  Many scripts are written using the
  Bash shell, and the main scripts are written using Perl.  Core
  programs needed to build the database and run the classifier are
  written in C++, and need to be compiled using g++.  Downloads of
  NCBI data are performed by wget and in some cases, by rsync.

  Finally, if you want to build your own database, you will need to
  install the Jellyfish k-mer counter, available at:

    http://www.cbcb.umd.edu/software/jellyfish/

  Note that Kraken only supports use of Jellyfish version 1.
  Jellyfish version 2 is not yet compatible with Kraken.

MiniKraken: To allow users with low-memory computing environments to
  use Kraken, we supply a reduced standard database that can be
  downloaded from the Kraken web site.  When Kraken is run with a
  reduced database, we call it MiniKraken.
  
  The database we make available is only 4 GB in size, and should
  run well on computers with as little as 8 GB of RAM.  Disk space
  for this database is also only 4 GB.


INSTALLATION
============

To begin using Kraken, you will first need to install it, and then
either download or create a database.

Kraken consists of two main scripts ("kraken" and "kraken-build"),
along with several programs and smaller scripts.  As part of the
installation process, all scripts and programs are installed in
the same directory.  After installation, you can move the main
scripts elsewhere, but moving the other scripts and programs
requires editing the scripts and changing the "$KRAKEN_DIR" variables.

Once a directory is selected, you need to run the following
commands in the directory where you extracted the Kraken
source (i.e., the directory where this README is located):

  ./install_kraken.sh $KRAKEN_DIR

(Replace "$KRAKEN_DIR" above with the directory where you want to
install Kraken's programs/directories.)

The install_kraken.sh script should compile all of Kraken's code
and setup your Kraken data directory.  Installation is successful
if you see the message "Kraken installation complete."

Once installation is complete, you may want to copy the two main
Kraken scripts into a directory found in your PATH variable 
(e.g., "$HOME/bin"):

  cp $KRAKEN_DIR/bin/kraken $HOME/bin
  cp $KRAKEN_DIR/bin/kraken-build $HOME/bin

After installation, you're ready to either create or download a
database.


KRAKEN DATABASES
================

A Kraken database is a directory containing at least 4 files:

* database.kdb: Contains the k-mer to taxon mappings
* database.idx: Contains minimizer offset locations in database.kdb
* taxonomy/nodes.dmp: Taxonomy tree structure + ranks
* taxonomy/names.dmp: Taxonomy names

Other files may be present as part of the database build process.

In interacting with Kraken, you should not have to directly reference
any of these files, but rather simply provide the name of the directory
in which they are stored.  Kraken allows both the use of a standard
database as well as custom databases; these are described in the sections
STANDARD KRAKEN DATABASE and CUSTOM DATABASES below, respectively.


STANDARD KRAKEN DATABASE
========================

To create the standard Kraken database, you can use the following command:

  kraken-build --standard --db $DBNAME

(Replace "$DBNAME" above with your preferred database name/location.
Please note that the database will use approximately 150 GB of
disk space during creation.)

This will download NCBI taxonomic information, as well as the
complete genomes in RefSeq for the bacterial, archaeal, and
viral domains.  After downloading all this data, the build
process begins; this is the most time-consuming step.  If you
have multiple processing cores, you can run this process with
multiple threads, e.g.:

  kraken-build --standard --threads 24 --db $DBNAME

Using 24 threads on our computer with 252 GB of RAM, the build
process took approximately an hour and a half (steps with an asterisk
have multi-threading enabled) in December 2013:

  10m50s  *Step 1 (create set)
     n/a   Step 2 (reduce database, optional and skipped)
  24m10s  *Step 3 (sort set)
   1m28s   Step 4 (GI number to file map)
   1m10s   Step 5 (File to taxon map)
  47m08s  *Step 6 (set LCA values)
  ------
  84m47s   Total build time

Note that if any step (including the initial downloads) fails,
the build process will abort.  However, kraken-build will
produce checkpoints throughout the installation process, and
will restart the build at the last incomplete step if you
attempt to run the same command again on a partially-built
database.

To create a custom database, or to use a database from another
source, see CUSTOM DATABASES.

Notes for users with lower amounts of RAM:

1) If you encounter problems with Jellyfish not being able
to allocate enough memory on your system to run the build
process, you can supply a smaller hash size to Jellyfish
using kraken-build's --jellyfish-hash-size switch.  Each space
in the hash table uses approximately 6.9 bytes, so using
"--jellyfish-hash-size 6400M" will use a hash table size of
6.4 billion spaces and require 44.3 GB of RAM.

2) Kraken's build process will normally attempt to minimize
disk writing by allocating large blocks of RAM and operating
within them until data needs to be written to disk.  However,
this extra RAM usage may exceed your capacity.  In such cases,
you may want to use kraken-build's --work-on-disk switch.  This
will minimize the amount of RAM usage and cause Kraken's build
programs to perform most operations off of disk files.  This
switch can also be useful for people building on a ramdisk or
solid state drive.  Please note that working off of disk files
can be quite slow on some computers, causing builds to take
several days if not weeks.


CLASSIFICATION
==============

To classify a set of sequences (reads), use the "kraken" command:

  kraken --db $DBNAME seqs.fa

Output will be sent to standard output by default.

Note that to obtain optimum speeds, Kraken's database should be
loaded into RAM first.  This can be done through use of a ramdisk,
if you have superuser permissions.  Failing that, you can use
the "--preload" switch to kraken, e.g.:

  kraken --preload --db $DBNAME seqs.fa

The database files will be loaded before classification using this
switch.  See MEMORY USAGE AND EFFICIENCY for more information.

The kraken program allows several different options:

- Multithreading: Use the "--threads NUM" switch to use multiple
  threads.

- Quick operation: Rather than searching all k-mers in a sequence,
  stop classification after the first database hit; use "--quick"
  to enable this mode.  Note that "--min-hits" will allow you to
  require multiple hits before declaring a sequence classified,
  which can be especially useful with custom databases when testing
  to see if sequences either do or do not belong to a particular
  genome.

- Sequence filtering: Classified or unclassified sequences can be
  sent to a file for later processing, using the "--classified-out"
  and "--unclassified-out" switches, respectively.

- Output redirection: Output can be directed using standard shell
  redirection ("|" or ">"), or using the "--output" switch.

- FASTQ input: Input is normally expected to be in FASTA format, but
  you can classify FASTQ data using the "--fastq-input" switch.

- Compressed input: Kraken can handle gzip and bzip2 compressed
  files as input by specifying the proper switch of "--gzip-compressed"
  or "--bzip2-compressed".

- Input format auto-detection: If regular files are specified on
  the command line as input, Kraken will attempt to determine the
  format of your input prior to classification.  You can disable this
  by explicitly specifying "--fasta-input", "--fastq-input",
  "--gzip-compressed", and/or "--bzip2-compressed" as appropriate.
  Note that use of the character device file "/dev/fd/0" to read
  from standard input (aka "stdin") will NOT allow auto-detection.

To get a full list of options, use "kraken --help".


OUTPUT FORMAT
=============

Each sequence classified by Kraken results in a single line of
output.  Output lines contain five tab-delimited fields; from
left to right, they are:

  1) "C"/"U": one letter code indicating that the sequence was
     either classified or unclassified.
  2) The sequence ID, obtained from the FASTA/FASTQ header.
  3) The taxonomy ID Kraken used to label the sequence; this is
     0 if the sequence is unclassified.
  4) The length of the sequence in bp.
  5) A space-delimited list indicating the LCA mapping of each k-mer
     in the sequence.  For example, "562:13 561:4 A:31 0:1 562:3"
     would indicate that:
     - the first 13 k-mers mapped to taxonomy ID #562
     - the next 4 k-mers mapped to taxonomy ID #561
     - the next 31 k-mers contained an ambiguous nucleotide
     - the next k-mer was not in the database
     - the last 3 k-mers mapped to taxonomy ID #562


CUSTOM DATABASES
================

We realize the standard database may not suit everyone's needs.  Kraken
also allows creation of customized databases.

To build a custom database:

  1) Install a taxonomy.  Usually, you will just use the NCBI taxonomy,
     which you can easily download using:
    
       kraken-build --download-taxonomy --db $DBNAME

     This will download the GI number to taxon map, as well as the
     taxonomic name and tree information from NCBI.  These files can
     be found in $DBNAME/taxonomy/ .  If you need to modify the taxonomy,
     edits can be made to the names.dmp and nodes.dmp files in this directory;
     the gi_taxid_nucl.dmp file will also need to be updated appropriately.

  2) Install a genomic library.  Three sets of standard genomes are
     made easily available through kraken-build:

     - bacteria: RefSeq complete bacterial/archaeal genomes
     - viruses: RefSeq complete viral genomes
     - human: GRCh38 human genome

     To download and install any one of these, use the "--download-library"
     switch, e.g.:

       kraken-build --download-library bacteria --db $DBNAME

     Other genomes can also be added, but such genomes must meet certain
     requirements:
     - Each replicon must be in a separate FASTA file
     - Each FASTA file must contain a GI number in the header,
       to allow Kraken to lookup the correct taxa.

     In practice, this means replicons must be downloaded from NCBI.
     Replicons without assigned GI numbers will require modifying the
     replicons' FASTA files (to assign a "fake" GI number) and then
     modifying the $DBNAME/taxonomy/gi_taxid_nucl.dmp file to provide
     your desired taxonomy information.

     If your genomes meet the requirements above, then you can add each
     replicon to your database's genomic library using the "--add-to-library"
     switch, e.g.:

       kraken-build --add-to-library chr1.fa --db $DBNAME
       kraken-build --add-to-library chr2.fa --db $DBNAME

  3) Once your library is finalized, you need to build the database.
     Depending on your size requirements, you may want to adjust the
     k-mer and/or minimizer lengths from the defaults.  Except for some
     small bookkeeping fields, a Kraken database will use

       12D + 8(4^M)

     bytes, where D is the number of distinct k-mers in your library and
     M is the length (in bp) of the minimizers.  Although D does increase
     as k increases, it is impossible to know exactly how many distinct
     k-mers will exist in a library for a given k without actually
     performing the count.  By default, k = 31 and M = 15.

     The minimizers serve to keep k-mers that are adjacent in query
     sequences close to each other in the database, which allows
     Kraken to exploit the CPU cache.  Changing the value of M can
     significantly affect the speed of Kraken, and neither increasing
     or decreasing M will guarantee faster or slower speed.

     To build the database, you'll use the "--build" switch:

       kraken-build --build --db $DBNAME

     As noted above, you may want to also use any of "--threads",
     "--kmer-len", or "--minimizer-len" to adjust the database build
     time and/or final size.

  4) Shrinking the database: The "--shrink" task allows you to take
     an existing Kraken database and create a smaller MiniKraken database
     from it.  The use of this option removes a specified percentage of
     k-mer/taxon pairs to create a new, smaller database.  For example:

       kraken-build --shrink 75 --db $DBNAME --new-db minikraken

     This will create a new database named "minikraken" that is 75% the
     size of the original, by removing 25% of k-mer/taxon pairs.

     The "--shrink" task is only meant to be run on a completed database.
     However, if you know before you create a database that you will
     only be able to use a certain amount of memory, you can use the
     "--max-db-size" switch for the "--build" task to provide a maximum
     size (in GB) for the database.  This allows you to create a MiniKraken
     database without having to create a full Kraken database first.

  A full list of options for kraken-build can be obtained using
  "kraken-build --help".

  After building a database, if you want to reduce the disk usage of
  the database you can use kraken-build's "--clean" switch to remove
  all intermediate files from the database directory.


MEMORY USAGE AND EFFICIENCY
===========================

Kraken's execution requires many random accesses to a very large file.
To obtain maximal speed, these accesses need to be made as quickly as
possible.  This means that the database must be in physical memory
during execution.  Although we provide the "--preload" option to Kraken
for users who cannot use a ramdisk, the ramdisk is likely the simplest
option, and is well-suited for installations on computers where Kraken
is to be run a majority of the time.  In addition, using a ramdisk
allows the initial start-up of Kraken to be accomplished much more quickly.
If a ramdisk is used, the "--preload" switch should not be used.

We also note that in some cases, "--preload" may not be needed (or even
advisable).  If you know that your database is already in memory (for
example, if it has been recently read or unzipped, then it should be in
your operating system cache, which resides in physical memory), then there
is no need to perform this step.  We have noticed that in low-memory (~8 GB)
situations, preloading a MiniKraken DB is actually much slower than simply
using "cat minikraken/database.* > /dev/null".  The selection of the best way
to get the database into memory is dependent on several factors, including
your total amount of RAM, operating system, and current free memory.  For this
reason, you may need to experiment with your own setup to find a good solution
for you.

To create a ramdisk, you will need to have superuser (root) permission.
As root, you can use the following commands to create a ramdisk:

  mkdir /ramdisk
  mount -t ramfs none /ramdisk

Optionally, you may have a trusted user who you want to be able to copy
databases into this directory.  In that case, you'll need to make that
user the owner of the directory via chown.

To put the database on the ramdisk, simply copy the database directory
to the ramdisk directory:

  cp -a $DBNAME /ramdisk

And then you can use it with Kraken by specifying the database copy on
the ramdisk, e.g.:

  kraken --db /ramdisk/$DBNAME seqs.fa

Note that anything copied into a ramdisk will be deleted if the ramdisk
is unmounted or the computer is restarted, so make sure that you have a
copy of the database on a hard disk (or other non-volatile storage).


PAIRED READS
============

Kraken does not query k-mers containing ambiguous nucleotides (non-ACGT).
If you have paired reads, you can use this fact to your advantage and
increase Kraken's accuracy by concatenating the pairs together with a
single 'N' between the sequences.  Using the "--paired" option when
running kraken will automatically do this for you; simply specify the
two mate pair files on the command line.  We have found this to raise
sensitivity by about 3 percentage points over classifying the sequences
as single-end reads.


SAMPLE REPORTS
==============

To get an idea as to Kraken's results across an entire sample, we provide
the kraken-report script.  It is used like this:

  kraken-report --db $DBNAME kraken.output

Note that the database used must be the same as the one used to generate
the output file, or the report script may encounter problems.  Output is
sent to standard output.

The output of kraken-report is tab-delimited, with one line per taxon.
The fields of the output, from left-to-right, are as follows:

  1) Percentage of reads covered by the clade rooted at this taxon
  2) Number of reads covered by the clade rooted at this taxon
  3) Number of reads assigned directly to this taxon
  4) A rank code, indicating (U)nclassified, (D)omain, (K)ingdom,
     (P)hylum, (C)lass, (O)rder, (F)amily, (G)enus, or (S)pecies.
     All other ranks are simply '-'.
  5) NCBI taxonomy ID
  6) indented scientific name 

The scientific names are indented using spaces, according to the tree
structure specified by the taxonomy.

By default, taxa with no reads assigned to (or under) them will not have
any output produced.  However, if you wish to have all taxa displayed,
you can use the "--show-zeros" switch to do so.  This can be useful if
you are looking to do further downstream analysis of the reports, and
want to compare samples.  Sorting by the taxonomy ID (using "sort -nf5")
can provide a consistent line ordering between reports.

In addition, we also provide the program kraken-mpa-report; this program
provides output in a format similar to MetaPhlAn's tab-delimited output.
For kraken-mpa-report, multiple Kraken output files can be specified on
the command line and each will be treated as a separate sample.  For each
taxon at the standard ranks (from domain to species), the count of reads
in each sample assigned to any node in the clade rooted at that taxon is
displayed.  Output of kraken-mpa-report is sent to standard output.


UPGRADING DATABASES TO v0.10+
=============================

The minimizer ordering in Kraken versions prior to v0.10.0-beta was a
simple lexicographical ordering that provided a suboptimal distribution
of k-mers within the bins.  Ideally, the bin sizes would be uniform,
but simple lexicographical ordering creates a bias toward low-complexity
minimizers.  To resolve this, the ordering is now "scrambled" by XORing all
minimizers with a predefined constant to toggle half of each minimizer's
bits before sorting.  The more evenly distributed bins provide better
caching performance, but databases created in this way are not compatible
with earlier versions of Kraken.  Kraken versions from v0.10.0-beta up to
(but not including) v1.0 will support the use of the older databases, but
we nonetheless recommend one of the two following options:

  1) Build a new database.  This is the preferred option, as a newly-created
     database will have the latest genomes and NCBI taxonomy information.

  2) Re-sort an existing database.  If you have a custom database, you may
     want to simply reformat the database to provide you with Kraken's
     increased speed.  To do so, you'll need to do the following:

       kraken-build --upgrade --db $DBNAME

     (NOTE: the --threads switch is both valid and encouraged with this
     operation.)

     This command will NOT delete your existing $DBNAME/database.*
     files, but will simply rename them.  If you're satisfied with the new
     database's performance, then you can use kraken-build's "--clean"
     option to remove the old files and save space.

     Sorting the database is step 3 of the build process, so you should
     expect a database upgrade to take about as long as step 3 took when
     building the original database.

Note that the rest of Kraken v0.10.0-beta's speed improvements are available
without upgrading or changing your database.