touala/picopore

A tool to reduce the size of Oxford Nanopore Technologies' datasets without losing information

Author's Note

picopore is no longer under active development. Due to improvements in ONT's native HDF5 compression, lossless and deep-lossless compression no longer effectively reduce the size of nanopore files. picopore's raw compression may still be of interest to users, but is no longer being actively maintained.

Picopore v1.2.0

A tool for reducing the size of Oxford Nanopore Technologies' datasets without losing information.

If you find Picopore useful, please cite it at http://dx.doi.org/10.12688/f1000research.11022.1

Options:

• Raw compression: reduces footprint by removing event detection and basecall data, leaving only raw signal, configuration data and a choice of FASTQ data, basecall summary, both or neither;
• Lossless compression: reduces footprint without reducing the ability to use other nanopore tools by using HDF5's inbuilt gzip functionality; (NOTE: as of May 2017, Oxford Nanopore Technologies implemented all compression strategies used in Picopore's lossless compression. Recently basecalled files will therefore not benefit from this compression.)
• Deep lossless compression: reduces footprint without removing any data by indexing basecalled dataset to the event detection dataset. (NOTE: deep lossless compression will have the greatest impact on 2D datasets. Further work to implement 1D^2 compression is in progress.)

Author: Scott Gigante, Walter & Eliza Hall Institute of Medical Research. Contact: Email, Twitter

Installation

Install via pypi

The latest stable version of Picopore is available on PyPi. Install it using the following command:

pip install picopore

Install via conda

Picopore and dependencies could also be installed using conda.

conda install picopore -c bioconda -c conda-forge

Install from source

For the bleeding edge, clone and install from GitHub.

git clone https://www.github.com/scottgigante/picopore
cd picopore
python setup.py install

Currently, h5py is only available on Windows via conda.

Requirements

Picopore runs on Python 2.7, 3.4, 3.5, 3.6 or 3.7 with development headers (python-dev or similar).

Picopore requires h5repack from hdf5-tools, which can be downloaded from https://support.hdfgroup.org/downloads/index.html or using sudo apt-get install hdf5-tools or similar.

Picopore requires the following Python packages:

• h5py
• future
• watchdog (for real-time compression)

In addition, h5py requires HDF5 1.8.4 or later (libhdf5-dev or similar). Difficulties resolving dependencies of h5py can be resolved by installing from your package manager, using sudo apt-get install python-h5py or similar.

Usage

commands: picopore
          picopore-realtime      monitors a directory for new reads and compresses them in real time
          picopore-test          compresses to temporary files and checks that all datasets and attributes are equal (lossless modes only)
          picopore-rename        renames groups and datasets within FAST5 files

usage: picopore [-h] --mode {lossless,deep-lossless,raw} [--revert] [--fastq]
                [--summary] [--manual STR] [-v] [-y] [-t INT] [--prefix STR]
                [--skip-root] [--print-every INT]
                [input [input ...]]

positional arguments:
  input                 list of directories or fast5 files to shrink

optional arguments:
  -h, --help            show this help message and exit
  --mode {lossless,deep-lossless,raw}
                        choose compression mode
  --revert              reverts files to original size (lossless modes only)
  --fastq, --no-fastq   retain FASTQ data (raw mode only) (Default: --fastq)
  --summary, --no-summary
                        retain summary data (raw mode only) (Default: --no-
                        summary)
  --manual STR          manually remove only groups whose paths contain STR
                        (raw mode only, regular expressions permitted,
                        overrides defaults)
  -v, --version         show version number and exit
  -y                    skip confirm step
  -t INT, --threads INT
                        number of threads (Default: 1)
  --prefix STR          add prefix to output files to prevent overwrite
  --skip-root, --no-skip-root
                        ignore files in root input directories for albacore
                        realtime compression (Default: --no-skip-root)
  --print-every INT     print a dot every approximately INT files, or -1 to
                        silence (Default: 100)

It is necessary to choose one compression mode out of lossless, deep-lossless, and raw.

Note that only lossless and deep-lossless are options for --revert.

For --manual raw compression, the entire group path is used for matching. For example, you could use the command picopore --mode raw --manual 1D.*Events [...] to remove the groups /Analyses/Basecall_1D_000/BaseCalled_template/Events and /Analyses/Basecall_1D_000/BaseCalled_complement/Events.

Compression Modes

Picopore compression allows most nanopore tools to operate unimpeded. We provide a list of software tools which can operate on compressed files unimpeded, and the process required to recover the necessary data if this is not possible.

Functionality	Lossless	Deep Lossless	Raw	Raw --no-fastq
Metrichor	yes	picopore --revert	yes	yes
nanonetcall	yes	picopore --revert	yes	yes
poretools fastq	yes	picopore --revert	yes	nanonetcall / Metrichor
poRe printfastq	yes	picopore --revert	yes	nanonetcall / Metrichor
nanopolish consensus	yes	picopore --revert	nanonetcall / Metrichor	nanonetcall / Metrichor

FAQs

Why would I want to shrink my fast5 files?

Nanopore runs are big. Really big. Over a long period of time, the storage footprint of a Nanopore lab will increase to unsustainable levels.

A large proportion of the data stored in ONT's fast5 files is unnecessary for the average end-user; during the basecalling process, a large amount of intermediary data is generated, and for most users who simply need the FASTQ, this data is useless.

Picopore solves this problem. Without removing the raw signal or configuration data used for basecalling, Picopore removes the intermediary datasets to reduce the size of your Nanopore dataset.

Do I lose functionality when using Picopore?

Lossless

Lossless compression uses HDF5's builtin compression, so all existing fast5 tools will work seamlessly.

• Use case: power users who wish to reduce server storage footprint

Deep Lossless

Deep lossless compression modifies the structure of your fast5 file: any data extraction tools will not work until you run python picopore.py --revert --mode deep-lossless [input].

• Use case: power users who wish to reduce the size of their files during data transfer, or for long-term storage

Raw

Raw compression removes the "squiggle-space" data. For most users, this data is not critical; the only tools we know of which use the squiggle-space data are nanopolish, nanoraw and nanonettrain. If you do not intend on using these tools, your tools will work as before. If you do intend to use these tools, the raw signal is retained, and you can resubmit the files for basecalling to generate new squiggle-space data.

• Use case: end users who are only interested in using the FASTQ data
• Use case: power users running local basecalling with limited local disk space, who wish to use FASTQ immediately and will submit reads to Metrichor at a later date

Raw --no-fastq

Minimal compression removes all data not required to rerun basecalling on the fast5 files. This is only recommended for long-term storage, and requires files to be re-basecalled for any data to be retrieved.

• Use case: users storing historical runs for archive purposes, with no short-term plans to use these reads

Do I lose any data when using Picopore?

Technically yes, but nothing that cannot be recovered. In the case where you need to access the data which has been removed, you can regenerate it using either picopore (on lossless compression) or using any basecaller provided by ONT (for other methods.)

Note that, since ONT's base calling is continuously improving, the basecalls generated when re-basecalling your data may not be the same, but in fact higher quality than before. If it is important that you retain the squiggle-space of the original called sequence, it is recommended that you use a lossless compression method.