Repository files navigation

AlphaTims is an open-source Python package that provides fast accession and visualization of unprocessed LC-TIMS-Q-TOF data fromBruker’s timsTOF Pro instruments. It indexes the data such that it can easily be sliced along all five dimensions: LC, TIMS, QUADRUPOLE, TOF and DETECTOR. It was developed by theMann Labs at the Max Planck Institute of Biochemistry as a modular tool of theAlphaPept ecosystem. To enable all hyperlinks in this document, please view it atGitHub.

• AlphaTims
  • About
  • License
  • Installation
    • One-click GUI
    • Pip installer
    • Developer installer
    • Installation issues
  • Test data
    • Test sample
    • LC
    • DDA
    • DIA
  • Usage
    • GUI
    • CLI
    • Python and jupyter notebooks
    • Other tools
  • Performance
    • Speed
    • RAM
  • Troubleshooting
  • How it works
    • Bruker raw data
    • TimsTOF objects in Python
    • Slicing TimsTOF objects
  • Future perspectives
  • Citing AlphaTims
  • How to contribute
  • Changelog
    • 1.0.0
    • 0.3.2
    • 0.3.1
    • 0.3.0
    • 0.2.8
    • 0.2.7

High-resolution quadrupole time-of-flight (Q-TOF) tandem mass spectrometry can be coupled to several other analytical techniques such as liquid chromatography (LC) and trapped ion mobility spectrometry (TIMS). LC-TIMS-Q-TOF has gained considerable interest since the introduction of theParallel Accumulation–Serial Fragmentation (PASEF) method in both data-dependent (DDA) and data-independent acquisition (DIA). With this setup, ion intensity values are acquired as a function of the chromatographic retention time, ion mobility, quadrupole mass to charge and TOF mass to charge. As these five-dimensional data points are detected at GHz rates, datasets often contain billions of data points which makes them impractical and slow to access. Raw data are therefore frequently binned for faster data analysis or visualization. In contrast, AlphaTims is a Python package that provides fast accession and visualization of unprocessed raw data. By recognizing that all measurements are ultimately arrival times linked to intensity values, it constructs an efficient set of indices such that raw data can be interpreted as a sparse five-dimensional matrix. On a modern laptop, this indexing takes less than half a minute for raw datasets of more than two billion datapoints. Following this step, interactive visualization of the same dataset can also be done in milliseconds. AlphaTims is freely available, open-source and available on all major Operating Systems. It can be used with a graphical user interface (GUI), a command-line interface (CLI) or as a regular Python package.

AlphaTims was developed by theMann Labs at the Max Planck Institute of Biochemistry and is freely available with anApache License. Since AlphaTims uses Bruker libraries (available in thealphatims/ext folder) additionalthird-party licenses are applicable. External Python packages (available in therequirements folder) have their own licenses, which can be consulted on their respective websites.

AlphaTims can be installed and used on all major operating systems (Windows, macOS and Linux).There are three different types of installation possible:

• One-click GUI installer: Choose this installation if you only want the GUI and/or keep things as simple as possible.
• Pip installer: Choose this installation if you want to use AlphaTims as a Python package in an existing Python 3.8 environment (e.g. a Jupyter notebook). If needed, the GUI and CLI can be installed with pip as well.
• Developer installer: Choose this installation if you are familiar with CLI tools,conda and Python. This installation allows access to all available features of AlphaTims and even allows to modify its source code directly. Generally, the developer version of AlphaTims outperforms the precompiled versions which makes this the installation of choice for high-throughput experiments.
• Docker: Use this installation if you want to use a container based workflow. This is usefull to preserve a clean environment or when running multiple tools that might have conflicting dependencies.

IMPORTANT: While AlphaTims is mostly platform independent, some calibration functions requireBruker libraries which are only available on Windows and Linux.

The GUI of AlphaTims is a completely stand-alone tool that requires no knowledge of Python or CLI tools. Click on one of the links below to download the latest release for:

• Windows
• macOS
• Linux

IMPORTANT: Please refer to theGUI manual for detailed instructions on the installation, troubleshooting and usage of the stand-alone AlphaTims GUI.

Older releases remain available on therelease page, but no backwards compatibility is guaranteed.

AlphaTims can be installed in an existing Python 3.8 environment with a singlebash command.Thisbash command can also be run directly from within a Jupyter notebook by prepending it with a!. The lightweight version of AlphaTims that purely focuses on data accession (no plotting without additional packages) can be installed with:

pip install alphatims

Installing AlphaTims like this avoids conflicts when integrating it in other tools, as this does not enforce strict versioning of dependancies. However, if new versions of dependancies are released, they are not guaranteed to be fully compatible with AlphaTims. While this should only occur in rare cases where dependencies are not backwards compatible, you can always force AlphaTims to use dependancy versions which are known to be compatible with:

pip install"alphatims[stable]"

NOTE: You might need to runpip install pip==21.0 before installing AlphaTims like this. Also note the double quotes".

Alternatively, some basic plotting functions can be installed with the following command:

pip install"alphatims[plotting]"

While the above command does allow usage of the full GUI, there are some known compatability issues with newer versions of bokeh. As such, it is generally advised to not use loose plotting dependancies and force a stable installation with:

pip install"alphatims[plotting-stable]"

When older samples need to be analyzed, it might be essential to install thelegacy version as well (See also thetroubleshooting section):

pip install"alphatims[legacy]"

When a new version of AlphaTims becomes available, the old version can easily be upgraded by running e.g. the command again with an additional--upgrade flag:

pip install"alphatims[plotting,legacy,stable]" --upgrade

The following extra options are available:

• stable
• plotting
• plotting-stable
• legacy
• legacy-stable
• development
• development-stable

NOTE: Multiple dependancy packs can be installed by comma-separation. Note however that this only works without spaces!

AlphaTims can also be installed in editable (i.e. developer) mode with a fewbash commands. This allows to fully customize the software and even modify the source code to your specific needs. When an editable Python package is installed, its source code is stored in a transparent location of your choice. While optional, it is advised to first (create and) navigate to e.g. a general software folder:

mkdir~/folder/where/to/install/softwarecd~/folder/where/to/install/software

The following commands assume you do not perform any additionalcd commands anymore.

Next, download the AlphaTims repository from GitHub either directly or with agit command. This creates a new AlphaTims subfolder in your current directory.

git clone https://github.com/MannLabs/alphatims.git

For any Python package, it is highly recommended to use aconda virtual environment. The best way to install an editable version of AlphaTims is to use AlphaTims' pre-built conda development environment (note that the--force flag overwrites an already existing AlphaTims environment):

conda env create --force --name alphatims --file alphatims/misc/conda_development_environment.yamlconda activate alphatims

Alternatively, a new conda environment can manually be created or AlphaTims can be installed in an already existing environment.Note that dependancy conflicts can occur with already existing packages in the latter case! Once a conda environment is activated, AlphaTims and all itsdependancies need to be installed. To take advantage of all features and allow development (with the-e flag), this is best done by installing both theplotting dependencies anddevelopment dependencies instead of only thecore dependencies:

conda create -n alphatims python=3.8 -yconda activate alphatimspip install -e"./alphatims[plotting-stable,development]"

By using the editable flag-e, all modifications to the AlphaTimssource code folder are directly reflected when running AlphaTims. Note that the AlphaTims folder cannot be moved and/or renamed if an editable version is installed.

The following steps are optional, but make working with AlphaTims slightly more convenient:

• To avoid callingconda activate alphatims andconda deactivate every time AlphaTims is used, the binary execution (which still reflects all modifications to the source code) can be added as an alias. On linux and MacOS, this can be done with e.g.:

  conda activate alphatimsalphatims_bin="$(which alphatims)"echo"alias alphatims='"${alphatims_bin}"'">>~/.bashrcconda deactivate

  Whenzsh is the default terminal instead ofbash, replace~/.bashrc with~/.zshrc. On Windows, the commandwhere alphatims can be used to find the location of the binary executable. This path can then be (permanently) added to Windows' path variable.
• When using Jupyter notebooks and multiple conda environments direcly from the terminal, it is recommended toconda install nb_conda_kernels in the conda base environment. Hereafter, running ajupyter notebook from the conda base environment should have apython [conda env: alphatims] kernel available, in addition to all other conda kernels in which the commandconda install ipykernel was run.

(WIP)

docker pull ghcr://MannLabs/alphatims:latest

See the generaltroubleshooting section.

AlphaTims is compatible with bothddaPASEF anddiaPASEF.

A test sample of human cervical cancer cells (HeLa, S3, ATCC) is provided for AlphaTims. These cells were cultured in Dulbecco's modified Eagle's medium (all Life Technologies Ltd., UK). Subsequently, the cells were collected, washed, flash-frozen, and stored at -80 °C.Following the previously publishedin-StageTip protocol, cell lysis, reduction, and alkylation with chloroacetamide were carried out simultaneously in a lysis buffer (PreOmics, Germany). The resultant dried peptides were reconstituted in water comprising 2 vol% acetonitrile and 0.1% vol% trifluoroacetic acid, yielding a 200 ng/µL solution. This solution was further diluted with water containing 0.1% vol% formic acid. The manufacturer's instructions were followed to load approximately 200ng peptides onto Evotips (Evosep, Denmark).

Single-run LC-MS analysis was executed via anEvosep One LC system (Evosep). This was coupled online with a hybridTIMS quadrupole TOF mass spectrometer (Bruker timsTOF Pro, Germany). A silica emitter (Bruker) was placed inside a nano-electrospray ion source (Captive spray source, Bruker) and connected to an 8 cm x 150 µm reverse phase column to perform LC. The column was packed with 1.5 µm C18-beads (Pepsep, Denmark). Mobile phases were water and acetonitrile, buffered with 0.1% formic acid. The samples were separated with a predefined 60 samples per day method (Evosep).

A ddaPASEF dataset is available fordownload from the release page. Each topN acquisition cycle consisted of 10 PASEF MS/MS scans, and the accumulation and ramp times were set to 100 ms. Single-charged precursors were excluded using a polygon filter in the m/z-ion mobility plane. Furthermore, all precursors, which reached the target value of 20000, were excluded for 0.4 min from the acquisition. Precursors were isolated with a quadrupole window of 2 Th for m/z <700 and 3 Th for m/z >700.

The same sample was acquired with diaPASEF and is also available fordownload from the release page. The "high-speed" method (mass range: m/z 400 to 1000, 1/K0: 0.6 – 1.6 Vs cm- 2, diaPASEF windows: 8 x 25 Th) was used, as described inMeier et al.

There are three ways to use AlphaTims:

• GUI: This allows to interactively browse, visualize and export the data.
• CLI: This allows to incorporate AlphaTims in automated workflows.
• Python: This allows to access data and explore it interactively with custom code.

NOTE: The first time you use a fresh installation of AlphaTims, it is often quite slow because some functions might still need compilation on your local operating system and architecture. Subsequent use should be a lot faster.

Please refer to theGUI manual for detailed instructions on the installation, troubleshooting and usage of the stand-alone AlphaTims GUI.

If the GUI was not installed through a one-click GUI installer, it can be activate with the followingbash command:

alphatims gui

Note that this needs to be prepended with a! when you want to run this from within a Jupyter notebook. When the command is run directly from the command-line, make sure you use the right environment (activate it with e.g.conda activate alphatims or set an alias to the binary executable).

The CLI can be run with the following command (after activating theconda environment withconda activate alphatims or if an alias was set to the alphatims executable):

alphatims -h

It is possible to get help about each function and their (required) parameters by using the-h flag. For instance, the commandalphatims export hdf -h will produce the following output:

************************* AlphaTims 0.0.210310 *************************Usage: alphatims export hdf [OPTIONS] BRUKER_D_FOLDER  Export BRUKER_D_FOLDER as hdf file.Options:  --disable_overwrite            Disable overwriting of existing files.  --enable_compression           Enable compression of hdf files. If set, this                                 roughly halves files sizes (on-disk), at the                                 cost of taking 2-10 longer accession times.  -o, --output_folder DIRECTORY  A directory for all output (blank means                                 `input_file` root is used).  -l, --log_file PATH            Save all log data to a file (blank means                                 'log_[date].txt' with date format                                 yymmddhhmmss in 'log' folder of AlphaTims                                 directory).  [default: ]  -t, --threads INTEGER          The number of threads to use (0 means all,                                 negative means how many threads to leave                                 available).  [default: -1]  -s, --disable_log_stream       Disable streaming of log data.  -p, --parameter_file FILE      A .json file with (non-required) parameters                                 (blank means default parameters are used).                                 NOTE: Parameters defined herein override all                                 default and given CLI parameters.  -e, --export_parameters FILE   Save currently selected parameters to a                                 parameter file.  -h, --help                     Show this message and exit.

For this particular command, the lineUsage: alphatims export hdf [OPTIONS] BRUKER_D_FOLDER shows that you always need to provide a path to aBRUKER_D_FOLDER and that all other options are optional (indicated by the brackets in[OPTIONS]). Each option can be called with a double dash-- followed by a long name, while common options also can be called with a single dash- followed by their short name. It is indicated what type of parameter is expected, e.g. aDIRECTORY for--output_folder or nothing forenable/disable flags. Defaults are also shown and all parameters will be saved in a log file. Alternatively, all used parameters can be exported with the--export_parameters option and the non-required ones can be reused with the--parameter_file.

IMPORTANT: Please refer to theCLI manual for detailed instructions on the usage and troubleshooting of the stand-alone AlphaTims CLI.

AlphaTims can be imported as a Python package into any Python script or notebook with the commandimport alphatims. Documentation for all functions is available in theRead the Docs API.

A briefJupyter notebook tutorial on how to use the API is also present in thenbs folder. When running locally it provides interactive plots, which are not rendered on GitHub. Instead, they are available as individual html pages in thenbs folder.

• Initial exploration of Bruker TimsTOF data files can be done by opening the .tdf file in the .d folder with anSQL browser.
• HDF files can be explored withHDF Compass orHDFView.
• Annotating Bruker TimsTOF data files can be done withAlphaPept
• Visualization of identified Bruker TimsTOF data files can be done withAlphaViz

Performance can be measured in function ofspeed orRAM usage.

Typical time performance statistics on data in-/output and slicing of standardHeLa datasets are available in theperformance notebook. All result can be summarized as follows:

On average, RAM usage is twice the size of a raw Bruker .d folder. Since most .d folders have file sizes of less than 10 Gb, a modern computer with 32 Gb RAM suffices to explore most datasets with ease.

Common installation/usage issues include:

• Always make sure you have activated the AlphaTims environment withconda activate alphatims. If this fails, make sure you have installedconda and have created an AlphaTims environment withconda create -n alphatims python=3.8.
• Nogit command. Make suregit is installed. In a notebook!conda install git -y might work.
• Wrong Python version. AlphaTims is only guaranteed to be compatible with Python 3.8. You can check if you have the right version with the commandpython --version (or!python --version in a notebook). If not, reinstall the AlphaTims environment withconda create -n alphatims python=3.8.
• Dependancy conflicts/issues. Pip changed their dependancy resolver withpip version 20.3. Downgrading or upgrading pip to version 20.2 or 21.0 withpip install pip==20.2 orpip install pip==21.0 (before runningpip install alphatims) could solve dependancy conflicts.
• AlphaTims is not found. Make sure you use the right folder. Local folders are best called by prefixing them with./ (e.g.pip install "./alphatims"). On some systems, installation specifically requires (not) to use single quotes' around the AlphaTims folder, e.g.pip install "./alphatims[plotting-stable,development]".
• Modifications to the AlphaTims source code are not reflected. Make sure you use the-e flag when usingpip install -e alphatims.
• Numpy does not work properly. On Windows,numpy==1.19.4 has some issues. After installing AlphaTims, downgrade NumPy withpip install numpy==1.19.3.
• Exporting PNG images with the CLI or Python package might not work out-of-the-box. If a conda environment is used, this can be fixed by runningconda install -c conda-forge firefox geckodriver in the AlphaTims conda environment. Alternatively, a file can be exported as html and opened in a browser. From the browser there is asave as png button available.
• GUI does not open. In some cases this can be simply because of using an incompatible (default) browser. AlphaTims has been tested with Google Chrome and Mozilla Firefox. Windows IE and Windows Edge compatibility is not guaranteed.
• When older Bruker files need to be processed as well, thelegacy dependencies are also needed. However, note that this requiresMicrosoft Visual C++ to be manually installed (on Windows machines) prior to AlphaTims installation! To include the legacy dependencies, install AlphaTims withpip install "alphatims[legacy]" orpip install "alphatims[legacy]" --upgrade if already pre-installed.
• When installed throughpip, the GUI cannot be started. Make sure you install AlphaTims withpip install "alphatims[plotting-stable]" to include the GUI with stable dependancies. If this was done and it still fails to run the GUI, a possible fix might be to runpip install panel==0.10.3 after AlphaTims was installed.
• Some external libraries are missing. On some OS, there might be libraries missing. As an exmaple, the following error message might pop up:OSError: libgomp.so.1: cannot open shared object file: No such file or directory. This can be solved by installing those manually, e.g. on Linux:apt-get install libgomp1.

The basic workflow of AlphaTims looks as follows:

• Read data from aBruker.d folder.
• Convert data to aTimsTOF object in Python and optionally store them as a persistentHDF5 file.
• Use Python'sslicing mechanism to retrieve data from this object e.g. for visualisation.

Also checkout:

• Thepaper for a complete overview.
• Thepresentation atISCB for a brief video.

Bruker stores TimsTOF raw data in a.d folder. The two main files in this folder areanalysis.tdf andanalysis.tdf_bin.

Theanalysis.tdf file is an SQL database, in which all metadata are stored together with summarised information. This includes theFrames table, wherein information about each individual TIMS cycle is summarised including the retention time, the number of scans (i.e. a single TOF push is related to a single ion mobility value), the summed intensity and the total number of ions that have hit the detector. More details about individual scans of the frames are available in thePasefFrameMSMSInfo (for PASEF acquisition) orDiaFrameMsMsWindows (for diaPASEF acquisition) tables. This includes quadrupole and collision settings of the frame/scan combinations.

Theanalysis.tdf_bin file is a binary file that contains the number of detected ions per individual scan, all detector arrival times and their intensity values. These values are grouped and compressed per frame (i.e. TIMS cycle), thereby allowing fast appendage during online acquisition.

AlphaTims first reads relevant metadata from theanalysis.tdf SQL database and creates a Python object of thebruker.TimsTOF class. Next, AlphaTims reads the summary information from theFrames table and creates three empty arrays:

• An emptytof_indices array, in which all TOF arrival times of each individual detector hit will be stored. Its size is determined by summing the number of detector hits for all frames.
• An emptyintensities array of the same size, in which all intensity values of each individual detector hit will be stored.
• An emptytof_indptr array, that will store the number of detector hits per scan. Its size is equal to(frame_max_index + 1) * scans_max_index + 1. It includes one additional frame to compensate for the fact that Bruker arrays are 1-indexed, while Python uses 0-indexing. The final+1 is because this array will be converted to an offset array, similar to the index pointer array of acompressed sparse row matrix. Typical values arescans_max_index = 1000 andframe_max_index = gradient_length_in_seconds * 10, resulting in approximatelylen(tof_indptr) = 10000 * gradient_length_in_seconds.

After reading thePasefFrameMSMSInfo orDiaFrameMsMsWindows table from theanalysis.tdf SQL database, four arrays are created:

• Aquad_indptr array that indexes thetof_indptr array. Each element points to an index of thetof_indptr where the voltage on the quadrupole and collision cell is adjusted. For PASEF acquisitions, this is typically 20 times per MSMS frame (turning on and off a value for 10 precursor selections) and once per change from an MS (precursor) frame to an MSMS (fragment) frame. For diaPASEF, this is typically twice to 10 times per frame and with a repetitive pattern over the frame cycle. This results in an array of approximatelylen(quad_indptr) = 100 * gradient_length_in_seconds. As with thetof_indptr array, this array is converted to an offset array with size+1.
• Aquad_low_values array oflen(quad_indptr) - 1. This array stores the lower m/z boundary that is selected with the quadrupole. For precursors without quadrupole selection, this value is set to -1.
• Aquad_high_values array, similar toquad_low_values.
• Aprecursor_indices array oflen(quad_indptr) - 1. For PASEF this array stores the index of the selected precursor. For diaPASEF, this array stores theWindowGroup of the fragment frame. A value of 0 indicates an MS1 ion (i.e. precursor) without quadrupole selection.

After processing this summarising information from theanalysis.tdf SQL database, the actual raw data from theanalysis.tdf_bin binary file is read and stored in the emptytof_indices,intensities andtof_indptr arrays.

Finally, three arrays are defined that allow quick translation offrame_,scan_ andtof_indices tort_values,mobility_values andmz_values arrays.

• Thert_values array is read read directly from theFrames table inanalysis.tdf and has a length equal toframe_max_index + 1. Note that an empty zeroth frame withrt = 0 is created to make Python's 0-indexing compatible with Bruker's 1-indexing.
• Themobility_values array is defined by using the functiontims_scannum_to_oneoverk0 fromtimsdata.dll on the first frame and typically has a length of1000.
• Similarly, themz_values array is defined by using the functiontims_index_to_mz fromtimsdata.dll on the first frame. Typically this has a length of400000.

All these arrays can be loaded into memory, taking up roughly twice as much RAM as the.d folder on disk. This increase in RAM memory is mainly due to the compression used in theanalysis.tdf_bin file. The HDF5 file can also be compressed so that its size is roughly halved and thereby has the same size as the Bruker.d folder, but (de)compression reduces accession times by 3-6 fold.

Once a Python TimsTOF object is available, it can be loaded into memory for ultrafast accession. Accession of thedata object is done by simple Python slicing such as e.g.selected_ion_indices = data[frame_selection, scan_selection, quad_selection, tof_selection]. This slicing returns apd.DataFrame for subsequent analysis. The columns of this dataframe contain all information for all selected ions, i.e.frame,scan,precursor andtof indices andrt,mobility,quad_low,quad_high,mz andintensity values. See thetutorial jupyter notebook for usage examples.

• Detection of:
  • precursor and fragment ions
  • isotopic envelopes (i.e. features)
  • fragment clusters (i.e. pseudo MSMS spectra)

Check out thepaper.

If you like AlphaTims you can give us astar to boost our visibility! All direct contributions are also welcome. Feel free to post a newissue or clone the repository and create apull request with a new branch. For an even more interactive participation, check out thediscussions.For more information seethe Contributors License Agreement.

This document gathers information on how to develop and contribute to this project.

In order to have release notes automatically generated, changes need to be tagged with labels.The following labels are used (should be safe-explanatory):breaking-change,bug,enhancement.

This package uses a shared release process defined in thealphashared repository. Please see the instructionsthere

The following changes were introduced in the following versions of AlphaTims. Download the latest version in theinstallation section.

• FEAT: tempmmap for large arrays by default.

• FEAT: cli/gui allow bruker data as argument.
• FEAT/FIX: Polarity included in frame table.
• FIX: utils cleanup.
• FIX: utils issues.
• FEAT: by default use -1 threads in utils.
• FIX: disable cla check.

• FIX/FEAT: Intensity correction when ICC is used. Note that this is only for exported data, not for visualized data.
• FEAT: By default, hdf files are now mmapped, making them much faster to initially load and use virtual memory in favor of residual memory.

• FEAT: Introduction of global mz calibration.
• FEAT: Introduction of dia_cycle for diaPASEF.
• CHORE: Verified Python 3.9 compatibility.
• FEAT: Included option to open Bruker raw data when starting the GUI.
• FEAT: Provided hash for TimsTOF objects.
• FEAT: Filter push indices.
• CHORE: included stable and loose versions for all dependancies

• FIX: Ensure stable version for one click GUI.
• FIX: Do not require plotting dependancies for CLI export csv selection.
• FIX: Import of very old diaPASEF samples where the analysis.tdf file still looks like ddaPASEF.
• FIX: frame pointers of fragment_frame table.
• FEAT: Include visual report in performance notebook.
• FEAT: Include DIA 120 sample in performance tests.
• FEAT: Show performance in README.
• FIX: Move python-lzf dependancy (to decompress older Bruker files) to legacy requirements, as pip install on Windows requires visual c++ otherwise.
• DOCS: BioRxiv paper link.
• FEAT/FIX: RT in min column.
• FEAT: CLI manual.
• FEAT: Inclusion of more coordinates in CLI.

• CHORE: Introduction of changelog.
• CHORE: Automated publish_and_release action to parse version numbers.
• FEAT/FIX: Include average precursor mz in MGF titles and set unknown precursor charges to 0.
• FIX: Properly resolve set_global argument ofalphatims.utils.set_threads.
• FIX: Set nogil option foralphatims.bruker.indptr_lookup.
• DOCS: GUI Manual typos.
• FEAT: Include buttons to download test data and citation in GUI.
• FEAT: Include option for progress_callback in alphatims.utils.pjit.
• FIX/FEAT: Older samples with TimsCompressionType 1 can now also be read. This is at limited performance.
• FEAT: By default use loose versioning for the base dependancies. Stable dependancy versions can be enforced withpip install "alphatims[stable]". NOTE: This option is not guaranteed to be maintained. Future AlphaTims versions might opt for an intermediate solution with semi-strict dependancy versioning.