Poretools: a toolkit for analyzing nanopore sequence data

1DNA sequencing with biological nanopores was proposed almost 20 years ago. This approach relies on the direct electrical detection of single DNA strands in contact with an individual pore. Single molecule detection and the absence of a prior amplification step means that extremely long fragments can be sequenced without any loss in quality. In May 2014, Oxford Nanopore Technologies released MinION™, the first commercially available nanopore DNA sequencing device. MinION™ is noteworthy for its portability, size (around the same length as an iPhone™) and USB 3.0 connectivity, meaning it can be run on a standard Internet-connected laptop. Research groups throughout the world are actively evaluating this device for a broad range of applications. Sequencing with the MinION yields raw signals reflecting modulation of the ionic current at each pore by a DNA molecule. The resulting time-series of nanopore translocation, ‘events’, are base-called by proprietary software running as a cloud service. The resulting files for each sequenced read are stored in ‘FAST5’ format, an application of the HDF5 format. However, at present, no specific software is available to facilitate downstream analyses starting with this file format.

2We have developed poretools, an open-source software toolkit that addresses the pressing need for methods to manipulate the FAST5 format and permit explorations of the raw nanopore event data and the resulting DNA sequences. Poretools provides an extensive set of data analysis methods that operate directly on either a single FAST5 file or a set of files from one or more sequencing runs. A Python programming library is provided to facilitate access to the FAST5 file structure and enable other researchers to extend the tools and create new analytical methods. In the following sections, we summarize the functionality currently available in poretools.

3The most fundamental functionality provided by poretools is the ability to convert the output data resulting from a MinION run from HDF5/FAST5 format to either FASTA or FASTQ format to facilitate analyses with sequence alignment and/or assembly software. This is accomplished with the fasta and fastq commands in the poretools suite. poretools fasta /path/to/fast5/example.fast5poretools fastq /path/to/fast5/example.fast5

4poretools fasta /path/to/fast5/example.fast5

5poretools fastq /path/to/fast5/example.fast5

6At the time of writing, each MinION run generates individual HDF5/FAST5 files for each sequenced read. Consequently, there are often tens of thousands of individual files that must be stored for a single experiment. Poretools provides two different strategies for facilitating the analysis of such datasets. The first approach allows one to execute a poretools command on an entire directory of FAST5 files. poretools fastq /path/to/fast5/directory/

7poretools fastq /path/to/fast5/directory/

8Alternatively, we provide a utility to combine a set of HDF5/FAST5 files into a single TAR file. This allows an entire run to be archived into one file and once combined, all other poretools commands are able to operate on each HDF5/FAST5 file therein. For example: poretools combine -o run.tar /path/to/fast5/directory/poretools fastq run.tar

9poretools combine -o run.tar /path/to/fast5/directory/poretools fastq run.tar

10There is a need to visualize MinION™ run performance to assess its quality and troubleshoot different fragmentation and library preparation strategies. Poretools provides two utilties, hist and yield_plot, that characterize the fragment size distribution and display a collector’s curve of the overall sequencing yield, respectively. Example commands are provided here, with corresponding figures shown (Fig. 1A and B). poretools hist /path/to/fast5/directory/poretools yield_plot /path/to/fast5/directory/ As summarized in Table 1, poretools also provides several utilities for extracting the low-level details that led to each base called sequence (see Table 1 for details). In particular, the events utility reports the mean current observed for each nanopore translocation event, as well as the time (in milliseconds) of each event and the k-mer that was predicted to have occupied the nanopore during the event. The squiggle utility permits visualization of this information (Fig. 1C). The Oxford Nanopore base-calling software uses a Hidden Markov Model to predict a fragment’s sequence based on this event data. We anticipate that the events utility (and others) will help developers explore improved base-calling strategies.

11poretools hist /path/to/fast5/directory/

12poretools yield_plot /path/to/fast5/directory/

13The utilities provided in the poretools suite will inevitably prove to be insufficient for every analysis that a researcher wishes to conduct. Recognizing this, we have developed a Python programming interface that researchers can use to directly access the sequence data, the raw nanopore event data and other metadata (e.g. the flowcell and run identifiers) contained in one or more FAST5 files. To demonstrate of the Python interface, the following code reports the start time, the specific nanopore and the based-called sequence for each FAST5 file in a sequencing run. from poretools import Fast5FileSetfast5s = Fast5FileSet(’/path/to/fast5/files/’)for fast5 in fast5s: start = fast5.get_start_time porenum = fast5.get_channel_number fq = fast5.get_fastq print porenum, start, fq.seq, fq.qual fast5.close

14from poretools import Fast5FileSet

15fast5s = Fast5FileSet(’/path/to/fast5/files/’)

16for fast5 in fast5s:

17start = fast5.get_start_time

18porenum = fast5.get_channel_number

19fq = fast5.get_fastq

20print porenum, start, fq.seq, fq.qual

21fast5.close

22The poretools software helps solve pressing requirements for analysis of nanopore sequencing data. By focusing on the Python development environment and adopting expected interface conventions as popularized by other popular bioinformatics tools such as samtools and bedtools, we expect that users will be able to rapidly exploit the functionality offered by this software. We anticipate that other toolkits will become available written in other programming languages. Further efforts are required for downstream analysis for common tasks including alignment and de novo assembly of both event and base-called sequence data from this platform.