A nanopore consensus algorithm using a signal-level hidden Markov model.
The program requires libhdf5 and a compiler that supports C++11. Development of the code is performed using gcc-4.8. libhdf5 can be automatically installed by the Makefile if you do not have it already (see below).
You will need to run git clone --recursive https://github.com/jts/nanopolish.git to get the source code and submodules. You can then compile nanopolish by running:
make
If you do not already have libhdf5, you can tell nanopolish to download and build it locally in the source tree:
make libhdf5.install nanopolish
The pipeline is still a prototype so it is fragile at the moment. It will be revised for general use after we submit the paper.
The reads that are input into the HMM must be output as a .fa file by poretools. This is important as poretools writes the path to the original .fast5 file (containing the signal data) in the fasta header. These paths must be correct or nanopolish cannot find the events for each read. Let's say you have exported your reads to reads.fa and you want to polish draft.fa. You should run:
make -f consensus.make READS=reads.fa ASSEMBLY=draft.fa
This will map the reads to the assembly with bwa mem -x ont2d and export a file mapping read names to fast5 files.
You can then run nanopolish consensus. It is recommended that you run this in parallel.
python nanopolish_makerange.py draft.fa | parallel --results nanopolish.results -P 8 nanopolish consensus -o nanopolish.{1}.fa -w {1} --r reads.pp.fa -b reads.pp.sorted.bam -g draft.fa -t 4
This command will run the consensus algorithm on eight 100kbp segments of the genome at a time, using 4 threads each. Change the -P and --threads options as appropriate for the machines you have available.
After all polishing jobs are complete, you can merge the individual segments together into the final assembly:
python nanopolish_merge.py draft.fa nanopolish.*.fa > polished.fa
If you have extremely high depth (for example you sequenced a virus) then poa and the hmm will take a very long time to run. I suggest downsampling to reasonable coverage before trying to call the consensus sequence.
The fast table-driven logsum implementation was provided by Sean Eddy as public domain code. This code was originally part of hmmer3.