MEGAHIT is a single node assembler for large and complex metagenomics NGS reads, such as soil. It makes use of succinct de Bruijn graph (SdBG) to achieve low memory assembly. The graph construction algorithm can self-adjust to use all available or moderate memory, and can be accelerated if a CUDA-enable GPU is provided. The GPU-accelerated version of MEGAHIT has been tested on NVIDIA GTX680 (4G memory) and Tesla K40c (12G memory).
MEGAHIT is suitable for 64-bit Linux and MAC OS X. It requires zlib, python 2.6 or greater and G++ 4.4 or greater (with -std=c++0x and OpenMP support). The GPU version further requires CUDA 5.5 or greater.
git clone https://github.com/voutcn/megahit.git
cd megahit
make
Notably, for MAC OS X, the g++ in the path is probably the sym-link of clang, which do not support OpenMP. Users should have the "real" G++ installed and use make CXX=/PATH/TO/G++ to specify the compiler.
Please refer to the release page.
If MEGAHIT is successfully compiled, it can be run by the following command:
./megahit [options] -m <max_memory_to_use> -l <max_read_len> {-r <reads.fa> | --input_cmd <command>}
or type make test in MEGAHIT's source directory for a quick test and ./megahit -h for the usage message.
To use the GPU version, run make use_gpu=1 to compile MEGAHIT, and run MEGAHIT with --use-gpu. GPU version has only been tested in Linux.
Users are requried to set a memory parameter -m for MEGAHIT. This parameter specifies the maximum memory that can be used by the SdBG constrution component of MEGAHIT. If 0 < -m < 1, the parameter specifies the fraction of the machine's total memory to be used; if -m >= 1, it specifies the memory in BYTE to be used. --mem-flag is another option for memory control.
Set the -m parameter to be 90-95% of the available memory (e.g. -m 0.9) and leave the --mem-flag default.
Please refer to this wiki page.
MEGAHIT accepts one fasta or fastq file as input. The input file can be gzip'ed. Alternatively, you can use the option --input-cmd to input reads from multiple files. Following the --input-cmd should be a command that outputs all reads to STDOUT in fasta or fastq format. A mix of fasta and fastq is also supported from version 0.2.0. Currently pair-end information is not used by MEGAHIT. Therefore pair-end files can be input to MEGAHIT as multiple single-end files. Some examples are shown on this wiki page.
###Choosing k
MEGAHIT uses multiple k-mer strategy. Minimum k, maximum k and the step for iteration can be set by options --k-min, --k-max and --k-step respectively. k must be odd numbers while the step must be an even number.
For ultra complex metagenomics data such as soil, a larger kmin, say 27, is recommended to reduce the complexity of the de Bruijn graph. Quality trimming is also recommended.
###Filtering (kmin+1)-mer
(kmin+1)-mer with multiplicity lower than d (default 2, specified by --min-count option) will be discarded. You should be cautious to set d less than 2, which will lead to a much larger and noisy graph. We recommend using the default value 2 for metagenomics assembly. If you want to use MEGAHIT to do generic assembly, please change this value according to the sequencing depth.
###Mercy k-mer
This is specially designed for metagenomics assembly to recover low coverage sequence. You can disable it with --no-mercy option.
If you have any questions or suggestions, please report an issue on Github.
- Li, D., Liu, C-M., Luo, R., Sadakane, K., and Lam, T-W., (2015) MEGAHIT: An ultra-fast single-node solution for large and complex metagenomics assembly via succinct de Bruijn graph. Bioinformatics, doi: 10.1093/bioinformatics/btv033 [PMID: 25609793].
MEGAHIT is released under GPLv3. Several third-party libs are used, including:
- CUB under "New BSD"" license
- kseq.h in klib under MIT license
- hash_{table, set, map}.h in IDBA package under GPLv2