methylprep is a python package for processing Illumina methylation array data. View on ReadTheDocs.

methylprep is part of a methyl-suite of python packages that provide functions to process and analyze DNA methylation data from Illumina arrays (27k, 450k, EPIC, and mouse arrays supported). The methylprep package contains functions for processing raw data files from arrays, or downloading and processing public data sets from GEO (the NIH Gene Expression Omnibus is a database repository), or from ArrayExpress. It contains both a command line interface (CLI) for processing data from local files, and a set of functions for building a custom pipeline in a jupyter notebook or python scripting environment. The aim is to offer a standard process, with flexibility for those who want it.

methylprep data processing has also been tested and benchmarked to match the outputs of two popular R packages: sesame (v1.10.4) and minfi (v1.38).

You should install all three components, as they work together. The parts include:

• methylprep: (this package) for processing idat files or downloading GEO datasets from NIH. Processing steps include

  • infer type-I channel switch
  • NOOB
  • poobah (p-value with out-of-band array hybridization, for filtering lose signal-to-noise probes)
  • qualityMask (to exclude historically less reliable probes)
  • nonlinear dye bias correction (AKA signal quantile normalization between red/green channels across a sample)
  • calculate beta-value, m-value, or copy-number matrix
  • large batch memory management, by splitting it up into smaller batches during processing

• methylcheck: for quality control (QC) and analysis, including

  • functions for filtering out unreliable probes, based on the published literature
    • Note that methylprep process will exclude a set of unreliable probes by default. You can disable that using the --no_quality_mask option from CLI.
  • sample outlier detection
  • array level QC plots, based on Genome Studio functions
  • a python clone of Illumina's Bead Array Controls Reporter software (QC)
  • data visualization functions based on seaborn and matplotlib graphic libraries.
  • predict sex of human samples from probes
  • interactive method for assigning samples to groups, based on array data, in a Jupyter notebook

• methylize provides more analysis and interpretation functions

  • differentially methylated probe statistics (between treatment and control samples)
  • volcano plots (which probes are the most different?)
  • manhattan plots (where in genome are the differences?)

methylprep maintains configuration files for your Python package manager of choice: pipenv or pip. Conda install is coming soon.

pip install methylprep

or if you want to install all three packages at once:

pip install methylsuite

The most common use case is processing .idat files on a computer within a command line interface. This can also be done in a Jupyter notebook, but large data sets take hours to run and Jupyter will take longer to run these than command line.

python -m methylprep -v process -d <filepath> --all

-d (data file path) is the only required option. The --all option applies the most common settings.

By default, these settings are designed to match the output of R's sesame processing. Prior to methylprep v1.4.0, the defaults matched minfi's output.

Here are some general options:

Use these options to specify file locations and array type:

Use these options to specify what gets saved from processing, and how it gets saved:

data_dir is the one required parameter. If you do not provide the file path for the project's sample_sheet CSV, it will find one based on the supplied data directory path. It will also auto detect the array type and download the corresponding manifest file for you.

Run the complete methylation processing pipeline for the given project directory, optionally exporting the results to file.

Returns: A collection of DataContainer objects for each processed sample

from methylprep import run_pipeline

data_containers = run_pipeline(data_dir, array_type=None, export=False, manifest_filepath=None, sample_sheet_filepath=None, sample_names=None)

Note: All the same input parameters from command line apply to run_pipeline, except --all. Type dir(methylprep.run_pipeline) in an interactive python session to see details.

Note: By default, if run_pipeline is called as a function in a script, a list of SampleDataContainer objects is returned. However, if you specify betas=True or m_value=True, a dataframe of beta values or m-values is returned instead. All methylcheck functions are designed to work on a dataframe or a folder to the processed data generated by run_pipeline.

This provides a scikit-learn style interface for you to specify every step in data processing by passing in a list of steps for each category of choices that will be encountered during idat file processing.

import methylprep
mysteps = ['all'] or ['infer_channel_switch', 'poobah', 'quality_mask', 'noob', 'dye_bias']
myexports = ['all'] or ['all', 'csv', 'poobah', 'meth', 'unmeth', 'noob_meth', 'noob_unmeth', 'sample_sheet_meta_data', 'mouse', 'control']
methylprep.make_pipeline(data_dir='.', steps=mysteps, exports=None, estimator='betas', **kwargs)

Use dir(methylprep.make_pipeline) for details.

If customizing the data processing steps interests you, you may also want to look at using the SampleDataContainer object, which is the output of processing when run in notebooks and beta_value or m_value is False. Each SampleDataContainer class object includes all of the sesame SigSet data sets and additional information about how the sample was processed.

methylprep provides a command line interface (CLI) so the package can be used directly in bash/batchfile or windows/cmd scripts as part of building your custom processing pipeline.

All invocations of the methylprep CLI will provide contextual help, supplying the possible arguments and/or options available based on the invoked command. If you specify verbose logging the package will emit log output of DEBUG levels and above.

python -m methylprep

usage: methylprep [-h] [-v] {process,sample_sheet} ...

Utility to process methylation data from Illumina IDAT files

positional arguments:
  {process,sample_sheet}
    process             process help
    sample_sheet        sample sheet help

optional arguments:
  -h, --help            show this help message and exit
  -v, --verbose         Enable verbose logging

The methylprep cli provides these top-level commands, which make it easier to use GEO datasets:

• process is the main function for processing methylation data from idat files. Covered already.
• download download and process public data sets in NIH GEO or ArrayExpress collections. Provide the public Accession ID and it will handle the rest.
• beta_bake combines download, meta_data, and file format conversion functions to produce a package that can be processed (with process) or loaded with methylcheck.load for analysis.
• sample_sheet will find/read/validate/create a sample sheet for a data set, or display its contents This is part of process and be applied using the --no_sample_sheet flag.
• alert scan GEO database and construct a CSV / dataframe of sample meta data and phenotypes for all studies matching a keyword
• composite download a bunch of datasets from a list of GEO ids, process them all, and combine into a large dataset
• meta_data will download just the meta data for a GEO dataset and convert it to a samplesheet CSV

There are thousands of publically accessible DNA methylation data sets available via the GEO (US NCBI NIH) https://www.ncbi.nlm.nih.gov/geo/ and ArrayExpress (UK) https://www.ebi.ac.uk/arrayexpress/ websites. This function makes it easy to import them and build a reference library of methylation data.

The CLI now includes a download option. Supply the GEO ID or ArrayExpress ID and it will locate the files, download the idats, process them, and build a dataframe of the associated meta data. This dataframe format should be compatible with methylcheck and methylize.

When processing large batches of raw .idat files, specify --batch_size to break the processing up into smaller batches so the computer's memory won't overload. This is off by default when using process but is ON when using download and set to batch_size of 100. Set to 0 to force processing everything as one batch. The output files will be split into multiple files afterwards, and you can recomine them using methylcheck.load.

Covered under Public GEO Datasets.

Find and parse the sample sheet in a given directory and emit the details of each sample. This is not required for actually processing data.

>>> python -m methylprep sample_sheet

usage: methylprep sample_sheet -d DATA_DIR

optional arguments:

~/methylprep$ python -m methylprep -v sample_sheet -d ~/GSE133062/GSE133062 --create
INFO:methylprep.files.sample_sheets:[!] Created sample sheet: ~/GSE133062/GSE133062/samplesheet.csv with 70 GSM_IDs
INFO:methylprep.files.sample_sheets:Searching for sample_sheet in ~/GSE133062/GSE133062
INFO:methylprep.files.sample_sheets:Found sample sheet file: ~/GSE133062/GSE133062/samplesheet.csv
INFO:methylprep.files.sample_sheets:Parsing sample_sheet
200861170112_R01C01
200882160083_R03C01
200861170067_R02C01
200498360027_R04C01
200498360027_R08C01
200861170067_R01C01
200861170072_R05C01
200498360027_R06C01
200861170072_R01C01
200861170067_R03C01
200882160070_R02C01
...

A tool to build a data set from a list of public datasets.

optional arguments:

Function to check for new datasets on GEO and update a csv each time it is run. Usable as a weekly cron command line function. Saves data to a local csv to compare with old datasets in _meta.csv. Saves the dates of each dataset from GEO; calculates any new ones as new rows. updates csv.

optional arguments:

Provides a more feature-rich meta data parser for public MINiML (formatted) GEO datasets. Run this after downloading the dataset using download command. This reads all the meta data from MINiML into a samplesheet.csv and meta data dataframe.

You can use meta_data to identify 'control' or samples containing a specific keyword (e.g. blood, tumor, etc) and remove any samples from sheet that lack these criteria, and delete the associated idats that don't have these keywords. After, run process on the rest, saving time. You can effectively ignore the parts of datasets that you don't need based on the associated meta data.

optional arguments:

                    Directory to search for MINiML file.

-c, --control |str| [experimental]: If flagged, this will look at the sample sheet and only save samples that appear to be"controls". -k KEYWORD, --keyword KEYWORD |str| [experimental]: Retain samples that include this keyword (e.g. blood, case insensitive) somewhere in samplesheet values. -s, --sync_idats | bool | [experimental]: If flagged, this will scan the data_dir and remove all idat files that are not in the filtered samplesheet, so they won't be processed. -o, --dont_download | bool | By default, this will first look at the local filepath (--data-dir) for GSE..._family.xml files. IF this is specified, it wont later look online to download the file. Sometimes a series has multiple files and it is easier to download, extract, and point this parser to each file instead.