meryl(1) in- and out-of-core kmer counting and utilities


Estimating memory requirements

meryl -P -m kmersize [-c #] [-p] -s seq.fasta

meryl -P -m kmersize [-c #] [-p] -n mercount

Building a table

meryl -B -m kmersize [-c #] [-p] [-v] [-f|-r|-C] [-L minoccurrence] [-U maxoccurrence] [-threads n | {-segments segments | -memory megabytes} [-configbatch [-sge jobname]]] -s seq.fasta -o tblprefix

meryl -countbatch number [-sgebuild "qsuboptionstring"] -o tblprefix

meryl -mergebatch number [-sgemerge "qsuboptionstring"] -o tblprefix

Performing operations on a table

meryl -M operation [-v] -s tblprefix [-s  tblprefix2 ...] -o output

Dumping a table

meryl -Dh -s tblprefix

meryl -Dt -n mincount -s tblprefix


meryl computes the kmer content of genomic sequences. Kmer content is represented as a list of kmers and the number of times each occurs in the input sequences. The kmer can be restricted to only the forward kmer, only the reverse kmer, or the canonical kmer (lexicographically smaller of the forward and reverse kmer at each location). Meryl can report the histogram of counts, the list of kmers and their counts, or can perform mathematical and set operations on the processed data files.

The output of meryl is two binary files, called a meryl database, which can be quickly dumped to provide a histogram of counts, or the actual counts. A C++ library is supplied for direct access to the files.


Estimate memory requirements. Given a sequence file (-s) or an upper limit on the number of mers in the file (-n), compute the table size (-t in build) to minimize the memory usage. This mode recognizes the following options:
-m #
size of a mer (required)
-c #
homopolymer compression (optional)
enable positions
-s seq.fasta
Sequence file to be scanned to determine the number of mers
-n #
compute params assuming file with this many mers in it

Only one of -s, -n need to be specified. If both are given, -s takes priority.

Compute the mer-count tables given a sequence file (-s) and lots of parameters. By default, both strands are processed.
only build for the forward strand
only build for the reverse strand
use canonical mers (assumes both strands)
-L #
DON'T save mers that occur less than # times
-U #
DON'T save mers that occur more than # times
-m #
size of a mer (required)
-c #
homopolymer compression (optional)
enable positions
-s seq.fasta
sequence to build the table for
-o tblprefix
output table prefix
entertain the user

The meryl process can run in one large memory batch, in many small memory batches, or under SGE control, all with or without using multiple CPU cores. By default, the computation is done as one large sequential process. Multi-threaded operation is possible, at additional memory expense, as is segmented operation, at additional I/O expense.

Threaded operation
Split the counting in to n almost-equally sized pieces. This uses an extra h MB (from -P) per thread.
-threads n
use n threads to build
Segmented, sequential operation
Split the counting into pieces that will fit into no more than m MB of memory, or into n equal sized pieces. Each piece is computed sequentially, and the results are merged at the end. Only one of -memory and -segments is needed.
-memory m
use at most m MB of memory per segment
-segments n
use n segments

Segmented, batched operation
Same as sequential, except this allows each segment to be manually executed in parallel. Only one of -memory and -segments is needed. Also see the EXAMPLE section on this page.
-memory m
use at most m MB of memory per segment
-segments n
use n segments
create the batches
-countbatch n
run batch number n
merge the batches

Batched mode can run on the grid.

-sge jobname
unique job name for this execution. Meryl will submit jobs with name mpjobname, ncjobname, nmjobname, for phases prepare, count and merge.
-sgebuild options
-sgemerge options
any additional options to qsub(1) (e.g., "-p -153 -pe thread 2 -A merylaccount") N.B. - -N will be ignored N.B. - be sure to quote the options
Given a list of tables, perform a math, logical or threshold operation. Unless specified, all operations take any number of databases. Math operations are:
count is the minimum count for all databases. If the mer does NOT exist in all databases, the mer has a zero count, and is NOT in the output.
count is the minimum count for all databases that contain the mer
count is the maximum count for all databases
count is sum of the counts for all databases
count is the first minus the second (binary only)
count is the absolute value of the first minus the second (binary only)

Logical operations are:

outputs mer iff it exists in all databases
outputs mer iff it exists in at least one, but not all, databases
outputs mer iff it exists in at least one database
outputs mer iff it exists in an odd number of databases

Threshold operations are:

lessthan x
outputs mer iff it has count < x
lessthanorequal x
outputs mer iff it has count <= x
greaterthan x
outputs mer iff it has count > x
greaterthanorequal x
outputs mer iff it has count >= x
equal x
outputs mer iff it has count == x

Threshold operations work on exactly one database.

-s tblprefix
use tblprefix as a database
-o tblprefix
create this output
entertain the user
Dump table (not all of these work)
Dump a histogram of the distance between the same mers.
Dump mers >= a threshold. Use -n to specify the threshold.
Count the number of mers, distinct mers and unique mers.
Dump (to stdout) a histogram of mer counts.
Read the count table from here (leave off the .mcdat or .mcidx).


Batch creation of a table

Initialize the compute with -configbatch, which needs all the build options. Execute all -countbatch jobs, then -mergebatch to complete.

meryl -configbatch -B [options] -o file meryl -countbatch 0 -o file meryl -countbatch 1 -o file ... meryl -countbatch N -o file meryl -mergebatch N -o file