FUNPACK

FUNPACK - the FMRIB UKBioBank Normalisation, Parsing And Cleaning Kit

FUNPACK is a Python library for pre-processing of UK BioBank data.

FUNPACK is developed at the Wellcome Centre for Integrative Neuroimaging (WIN@FMRIB), University of Oxford. FUNPACK is in no way endorsed, sanctioned, or validated by the UK BioBank.

FUNPACK comes bundled with metadata about the variables present in UK BioBank data sets. This metadata can be obtained from the UK BioBank online data showcase

Installation

Install FUNPACK from conda-forge:

conda install -c conda-forge fmrib-unpack

Or using pip:

pip install fmrib-unpack

The FUNPACK source code can be found at https://git.fmrib.ox.ac.uk/fsl/funpack/.

Introductory notebook

The funpack_demo command will start a Jupyter Notebook which introduces the main features provided by FUNPACK. A non-interactive version of this notebook can be found at https://open.win.ox.ac.uk/pages/fsl/funpack/demo.html.

If you are using pip, you need to install a few additional dependencies:

pip install fmrib-unpack[demo]

You can then start the demo by running fmrib_unpack_demo.

Note

The introductory notebook uses bash, so is unlikely to work on Windows.

Usage

General usage is as follows:

fmrib_unpack [options] output.tsv input1.tsv input2.tsv

You can get information on all of the options by typing fmrib_unpack --help.

The fmrib_unpack command was called funpack in older versions of FUNPACK, but was changed to fmrib_unpack in 3.0.0 to avoid a naming conflict with an unrelated software package.

Options can be specified on the command line, and/or stored in a configuration file. For example, the options in the following command line:

fmrib_unpack \
  --overwrite \
  --write_log \
  --icd10_map_file icd_codes.tsv \
  --category 10 \
  --category 11 \
  output.tsv input1.tsv input2.tsv

Could be stored in a configuration file config.txt:

overwrite
write_log
icd10_map_file icd_codes.tsv
category       10
category       11

And then executed as follows:

fmrib_unpack -cfg config.txt output.tsv input1.tsv input2.tsv

Features

FUNPACK allows you to perform various data sanitisation and processing steps on your data, such as:

• NA value replacement: Specific values for some columns can be replaced with NA, for example, variables where a value of -1 indicates Do not know.

• Categorical recoding: Certain categorical columns can re-coded. For example, variables where a value of 555 represents half can be recoded so that 555 is replaced with 0.5.

• Child value replacement: NA values within some columns which are dependent upon other columns may have values inserted based on the values of their parent columns.

See the introductory notebook for a more comprehensive overview of the features available in FUNPACK.

Built-in rules

FUNPACK contains a large number of built-in rules which have been specifically written to pre-process UK BioBank data variables. These rules are stored in the following files [*]:

• funpack/configs/fmrib/datacodings_*.tsv: Cleaning rules for data codings

• funpack/configs/fmrib/variables_*.tsv: Cleaning rules for individual variables

• funpack/configs/fmrib/processing.tsv: Processing steps

• funpack/configs/fmrib/categories.tsv: Variable categories

[*]

The funpack/ directory is located in <python-env>/lib/python<X.Y>/site-packages/funpack/, where <python-env> is the location of your Python environment directory, and <X.Y> is the Python version you are using.

You can use these rules by using the FMRIB configuration profile:

fmrib_unpack -cfg fmrib output.tsv input.tsv

You can customise or replace these files as you see fit. You can also pass your own versions of these files to FUNPACK via the --variable_file, --datacoding_file, --type_file, --processing_file, and --category_file command-line options respectively. FUNPACK will load all variable and datacoding files, and merge them into a single table which contains the cleaning rules for each variable.

Note

The fmrib configuration profile is managed and released separately from FUNPACK at https://git.fmrib.ox.ac.uk/fsl/funpack-fmrib-config/. However, it is automatically installed alongside FUNPACK, so if you have FUNPACK, you can use the fmrib profile. If you are using FUNPACK from a source checkout, you may need to manually install the fmrib-unpack-fmrib-config package from PyPi or conda-forge.

Creating your own rule files

To define rules at the data-coding level, create one or more .tsv files with an ID column containing the data-coding ID, and any of the following columns:

• NAValues: A comma-separated list of values to replace with NA

• RawLevels A comma-separated list of values to be replaced with corresponding values in NewLevels.

• NewLevels A comma-separated list of replacement values for each of the values listed in RawLevels.

To apply these rules, pass your .tsv file(s) to funpack with the --datacoding_file option. They will be applied to all variables which use the data-coding(s) listed in the file(s).

To define rules at the variable level, create one or more .tsv files with an ID column containing the variable ID, and any of the following columns:

• NAValues: As above

• RawLevels As above

• NewLevels As above

• ParentValues: A comma-separated list of expressions on parent variables, defining conditions which should trigger child-value replacement.

• ChildValues: A comma-separated list of values to insert into the variable when the corresponding expression in ParentValues evaluates to true.

• Clean: A comma-separated list of cleaning functions to apply to the variable.

Output

The main output of FUNPACK is a plain-text file [†] which contains the input data, after cleaning and processing, potentially with some columns removed, and new columns added.

If you used the --suppress_non_numerics option, the main output file will only contain the numeric columns. You can combine this with the --write_non_numerics option to save non-numeric columns to a separate file.

You can use any tool of your choice to load this output file, such as Python, MATLAB, or Excel. It is also possible to pass the output back into FUNPACK.

[†]

If your output file name ends with .csv, the file will be comma-separated, and if your output file name ends with .tsv, the file will be tab-separated.

Loading output into MATLAB

If you are using MATLAB, you have several options for loading the FUNPACK output. The best option is readtable, which will load column names, and will handle both non-numeric data and missing values. Use readtable like so (assuming that you generated a tab-separated file):

data = readtable('out.tsv', 'FileType', 'text');

The readtable function returns a table object, which stores each column as a separate vector (or cell-array for non-numeric columns). If you are only interested in numeric columns, you can retrieve them as an array like this:

data    = data(:, vartype('numeric'));
rawdata = data.Variables;

The readtable function will potentially rename the column names to ensure that they are are valid MATLAB identifiers. You can retrieve the original names from the table object like so:

colnames = data.Properties.VariableDescriptions';

If you have used the --write_description or --description_file options, you can load in the descriptions for each column as follows:

descs = readtable('out_descriptions.tsv', ...
                  'FileType', 'text', ...
                  'Delimiter', '\t',  ...
                  'ReadVariableNames',false);
descs = [descs; {'eid', 'ID'}];
idxs  = cellfun(@(x) find(strcmp(descs.Var1, x)), colnames, ...
                'UniformOutput', false);
idxs  = cell2mat(idxs);
descs = descs.Var2(idxs);

Tests

To run the test suite, you need to install some additional dependencies:

pip install fmrib-unpack[test]

Then you can run the test suite using pytest:

pytest

macOS issues

FUNPACK makes extensive use of the Python multiprocessing module to speed up certain steps in its processing pipeline. FUNPACK relies on the POSIX fork() mechanism, so that worker processes may inexpensively inherit the memory space of the main process (often referred to as copy-on-write). This is to avoid having to serialise the data set being processed (stored internally as a pandas.DataFrame).

In python 3.8 on macOS, the default method used by the multiprocessing module was changed from fork to spawn, due to changes in macOS 10.13 restricting the use of fork() for safety reasons. Some background information on this change can be found at https://bugs.python.org/issue33725, and at this blog post.

FUNPACK therefore explicitly sets the method used by the multiprocessing to fork, to take advantage of copy-on-write semantics. Using fork() on macOS should be safe for single-threaded parent processes, but as FUNPACK calls fork() numerous times (by creating and discarding multiprocessing.Pool() objects on an as-needed basis), this assumption may not be valid, and FUNPACK may crash with an error message resembling the following:

+[SomeClass initialize] may have been in progress in another thread
when fork() was called. We cannot safely call it or ignore it in the
fork() child process. Crashing instead.

You might be able to work around this error by setting an environment variable before calling FUNPACK, like so:

export OBJC_DISABLE_INITIALIZE_FORK_SAFETY=YES
fmrib_unpack ...

Citing

If you would like to cite FUNPACK, please refer to its Zenodo page.