timspy

TimsPy: access for raw timsTOF Pro data for data scientists


Keywords
timsTOFpro, numpy, data, science, mass, spectrometry
License
BSD-3-Clause
Install
pip install timspy==0.9.1

Documentation

TimsPy

A data science friendly data access to timsTOF Pro mass spectrometry data.

Requirements

In general, the software should work on Linux, Windows, or MacOS. Python3.6 or higher versions are tested.

On Windows, install Microsoft Visual Studio from here to make use of C++ or Python code. On Linux, have clang++ or g++ installed (clang is better). On macOS, install x-tools command line tools.

From terminal (assuming you have python and pip included in the system PATH) write

pip install timspy

For a direct installation from github:

pip install git+https://github.com/michalsta/opentims

On Windows: we have noticed issues with the numpy==1.19.4 due to changes in Intel's fmod function, unrelated to our work. If you keep on experiencing these issues, install numpy==1.19.3.

pip uninstall numpy
pip install numpy==1.19.3

For version supporting vaex and HDF5, write:

pip install timspy[vaex]

or simply add in the missing modules to the existing installation:

pip install vaex-core vaex-hdf5 h5py

What can you expect?

Simple way to get data out of results collected with your Bruker timsTOF Pro from Python. This definitely ain't no rocket science, but is pretty useful!

For example:

import pathlib
from pprint import pprint

from timspy.df import TimsPyDF
from timspy.dia import TimsPyDIA

path = pathlib.Path('path_to_your_data.d')
D = TimsPyDF(path) # get data handle
print(D)
# TimsPyDF(404183877 peaks)

print(len(D)) # The number of peaks.
# 404183877 

D.intensity_per_frame() # Return combined intensity for each frame.
# array([ 95955, 579402, 907089, ..., 406508,   8097,   8633])

try:
    import opentims_bruker_bridge
    all_columns = ('frame','scan','tof','intensity','mz','inv_ion_mobility','retention_time')
except ModuleNotFoundError:
    print("Without Bruker proprietary code we cannot yet perform tof-mz and scan-dt transformations.")
    print("Download 'opentims_bruker_bridge' if you are on Linux or Windows.")
    print("Otherwise, you will be able to use only these columns:")
    all_columns = ('frame','scan','tof','intensity','retention_time')


# We consider the following columns:
print(all_columns)
# ('frame', 'scan', 'tof', 'intensity', 'mz', 'inv_ion_mobility', 'retention_time')


# Get a dict with data from frames 1, 5, and 67.
pprint(D.query(frames=[1,5,67], columns=all_columns))
#         frame  scan     tof  intensity           mz  inv_ion_mobility  retention_time
# 0           1    33  312260          9  1174.655791          1.601142        0.326492
# 1           1    34  220720          9   733.480941          1.600000        0.326492
# 2           1    34  261438          9   916.952389          1.600000        0.326492
# 3           1    36   33072          9   152.355651          1.597716        0.326492
# 4           1    36  242110          9   827.311421          1.597716        0.326492
# ...       ...   ...     ...        ...          ...               ...             ...
# 224732     67   917  192745         51   619.285021          0.600774        7.405654
# 224733     67   917  201838         54   655.343937          0.600774        7.405654
# 224734     67   917  205954         19   672.001670          0.600774        7.405654
# 224735     67   917  236501         57   802.160552          0.600774        7.405654
# 224736     67   917  289480         95  1055.203750          0.600774        7.405654
# 
# [224737 rows x 7 columns]




# Get a dict with each 10th frame, starting from frame 2, finishing on frame 1000.   
pprint(D.query(frames=slice(2,1000,10), columns=all_columns))
#         frame  scan     tof  intensity           mz  inv_ion_mobility  retention_time
# 0           2    33   97298          9   302.347671          1.601142        0.434706
# 1           2    33  310524          9  1165.327281          1.601142        0.434706
# 2           2    34  127985          9   391.984100          1.600000        0.434706
# 3           2    35  280460          9  1009.675130          1.598858        0.434706
# 4           2    37  329377         72  1268.626208          1.596575        0.434706
# ...       ...   ...     ...        ...          ...               ...             ...
# 669552    992   909  198994          9   643.956206          0.609747      106.710279
# 669553    992   909  282616          9  1020.466272          0.609747      106.710279
# 669554    992   912  143270          9   440.966974          0.606382      106.710279
# 669555    992   915  309328          9  1158.922133          0.603017      106.710279
# 669556    992   916  224410          9   749.264705          0.601896      106.710279
# 
# [669557 rows x 7 columns]


# Get all MS1 frames 
# pprint(D.query(frames=D.ms1_frames, columns=all_columns))
# ATTENTION: that's quite a lot of data!!! You might exceed your RAM.


# If you want to extract not every possible columnt, but a subset, use the columns argument:
pprint(D.query(frames=slice(2,1000,10), columns=('tof','intensity',)))
#            tof  intensity
# 0        97298          9
# 1       310524          9
# 2       127985          9
# 3       280460          9
# 4       329377         72
# ...        ...        ...
# 669552  198994          9
# 669553  282616          9
# 669554  143270          9
# 669555  309328          9
# 669556  224410          9
# 
# [669557 rows x 2 columns]
# 
# This will reduce your memory usage.


# Still too much memory used up? You can also iterate over frames:
it = D.query_iter(slice(10,100,10), columns=all_columns)
pprint(next(it))
#        frame  scan     tof  intensity           mz  inv_ion_mobility  retention_time
# 0         10    34  171284          9   538.225728          1.600000        1.293682
# 1         10    36   31282          9   148.904423          1.597716        1.293682
# 2         10    38  135057          9   414.288925          1.595433        1.293682
# 3         10    39  135446          9   415.533724          1.594291        1.293682
# 4         10    41  188048          9   601.058384          1.592008        1.293682
# ...      ...   ...     ...        ...          ...               ...             ...
# 11470     10   908   86027        215   272.343029          0.610869        1.293682
# 11471     10   908  304306          9  1132.219605          0.610869        1.293682
# 11472     10   913  207422          9   677.993343          0.605260        1.293682
# 11473     10   916   92814         13   290.222999          0.601896        1.293682
# 11474     10   916   95769         86   298.185400          0.601896        1.293682
# 
# [11475 rows x 7 columns]


pprint(next(it))
#       frame  scan     tof  intensity           mz  inv_ion_mobility  retention_time
# 0        20    33  359979         31  1445.637778          1.601142        2.366103
# 1        20    33  371758         10  1516.851302          1.601142        2.366103
# 2        20    34  170678          9   536.019344          1.600000        2.366103
# 3        20    37  187676          9   599.626478          1.596575        2.366103
# 4        20    38   12946          9   115.828412          1.595433        2.366103
# ...     ...   ...     ...        ...          ...               ...             ...
# 6555     20   915   18954          9   126.209155          0.603017        2.366103
# 6556     20   915  136901        111   420.206274          0.603017        2.366103
# 6557     20   915  137327         26   421.579263          0.603017        2.366103
# 6558     20   915  137500          9   422.137478          0.603017        2.366103
# 6559     20   916   96488          9   300.139081          0.601896        2.366103
# 
# [6560 rows x 7 columns]



# All MS1 frames, but one at a time
iterator_over_MS1 = D.query_iter(D.ms1_frames, columns=all_columns)
pprint(next(it))
#        frame  scan     tof  intensity          mz  inv_ion_mobility  retention_time
# 0         30    33  210334          9  689.957428          1.601142        3.440561
# 1         30    34   24628          9  136.421775          1.600000        3.440561
# 2         30    34   86106          9  272.547880          1.600000        3.440561
# 3         30    34  165263          9  516.505046          1.600000        3.440561
# 4         30    35  224302          9  748.800355          1.598858        3.440561
# ...      ...   ...     ...        ...         ...               ...             ...
# 13076     30   903  266788          9  942.579660          0.616478        3.440561
# 13077     30   905  164502          9  513.791592          0.614234        3.440561
# 13078     30   914   77215        125  249.976426          0.604139        3.440561
# 13079     30   914   78321         22  252.731087          0.604139        3.440561
# 13080     30   917   92056         20  288.197894          0.600774        3.440561
# 
# [13081 rows x 7 columns]

pprint(next(it))
#       frame  scan     tof  intensity           mz  inv_ion_mobility  retention_time
# 0        40    33  310182          9  1163.493906          1.601142        4.511882
# 1        40    33  362121         11  1458.460526          1.601142        4.511882
# 2        40    34  364148         76  1470.646985          1.600000        4.511882
# 3        40    35  364858         40  1474.927542          1.598858        4.511882
# 4        40    37    4059          9   101.290007          1.596575        4.511882
# ...     ...   ...     ...        ...          ...               ...             ...
# 7354     40   915  126467        100   387.276839          0.603017        4.511882
# 7355     40   915  130612        106   400.197513          0.603017        4.511882
# 7356     40   916  131559         77   403.179227          0.601896        4.511882
# 7357     40   916  270542          9   960.772728          0.601896        4.511882
# 7358     40   917  127329         88   389.946380          0.600774        4.511882
# 
# [7359 rows x 7 columns]


# The frame lasts a convenient time unit that well suits chromatography peak elution.
# What if you were interested instead in finding out which frames eluted in a given time 
# time of the experiment?
# For this reasone, we have prepared a retention time based query:
# suppose you are interested in all frames corresponding to all that eluted between 10 and 12
# minute of the experiment.
# Well, here goes nothing:
D.rt_query(10,12)
#         frame  scan     tof  ...           mz  inv_ion_mobility  retention_time
# 0          92    33  361758  ...  1456.283499          1.601142       10.086899
# 1          92    36   65738  ...   222.282248          1.597716       10.086899
# 2          92    41  308330  ...  1153.590878          1.592008       10.086899
# 3          92    43  123618  ...   378.518961          1.589725       10.086899
# 4          92    48   65346  ...   221.365052          1.584017       10.086899
# ...       ...   ...     ...  ...          ...               ...             ...
# 128129    109   914  138760  ...   426.214230          0.604139       11.910014
# 128130    109   914  142129  ...   437.210911          0.604139       11.910014
# 128131    109   914  144566  ...   445.252770          0.604139       11.910014
# 128132    109   916  138933  ...   426.775504          0.601896       11.910014
# 128133    109   917  373182  ...  1525.576529          0.600774       11.910014

# [128134 rows x 7 columns]



# Get numpy array with raw data in a given range 1:10
pprint(D[1:10])
# array([[     1,     33, 312260,      9],
#        [     1,     34, 220720,      9],
#        [     1,     34, 261438,      9],
#        ...,
#        [     9,    913, 204042,     10],
#        [     9,    914, 358144,      9],
#        [     9,    915, 354086,      9]], dtype=uint32)


# Simple access to 'analysis.tdf'? Sure:
D.tables_names()
#                         name
# 0             GlobalMetadata
# 1        PropertyDefinitions
# 2             PropertyGroups
# 3            GroupProperties
# 4            FrameProperties
# 5              FrameMsMsInfo
# 6            CalibrationInfo
# 7                     Frames
# 8                   Segments
# 9                   ErrorLog
# 10             MzCalibration
# 11           TimsCalibration
# 12  DiaFrameMsMsWindowGroups
# 13       DiaFrameMsMsWindows
# 14          DiaFrameMsMsInfo

# Just choose a table now:
D.table2df('TimsCalibration')
#    Id  ModelType  C0   C1          C2         C3    C4  C5        C6          C7         C8           C9
# 0   1          2   1  917  213.599846  75.817289  33.0   1 -0.009066  135.436364  13.326079  1663.341457
# and there will be pandas all the way down. 

# Also, one table (Frames) is so important, that you can get it immediately: 
D.frames
#           Id         Time Polarity  ScanMode  ...  TimsCalibration  PropertyGroup  AccumulationTime  RampTime
# 0          1     0.326492        +         9  ...                1              1            99.953    99.953
# 1          2     0.434706        +         9  ...                1              1            99.953    99.953
# 2          3     0.540987        +         9  ...                1              1            99.953    99.953
# 3          4     0.648887        +         9  ...                1              1            99.953    99.953
# 4          5     0.756660        +         9  ...                1              1            99.953    99.953
# ...      ...          ...      ...       ...  ...              ...            ...               ...       ...
# 11548  11549  1243.494142        +         9  ...                1              1            99.953    99.953
# 11549  11550  1243.599171        +         9  ...                1              1            99.953    99.953
# 11550  11551  1243.707291        +         9  ...                1              1            99.953    99.953
# 11551  11552  1243.811222        +         9  ...                1              1            99.953    99.953
# 11552  11553  1243.919337        +         9  ...                1              1            99.953    99.953

# [11553 rows x 17 columns]

Basic plotting

You can plot the overview of your experiment. Continuing on the previous example:

D.plot_TIC()

D.plot_peak_counts()

intensities_matrix, mz_bin_borders, inv_ion_mobility_bin_borders = \
    D.intensity_given_mz_inv_ion_mobility()

D.plot_intensity_given_mz_inv_ion_mobility(
    intensities_matrix, 
    mz_bin_borders,
    inv_ion_mobility_bin_borders
)

DIA data

To get meaningful information out of the DIA experiments, you have to combine data from both analysis.tdf_raw and analysis.tdf. With TimsPy that is simple:

"""
We made a class to simplify access to DIA data.
Its parent class is TimsPyDF, so you can use all of its methods.
"""
from timspy.dia import TimsPyDIA



path = 'path/to/your/data.d'
D = TimsPyDF(path) # get data handle

# Interested in Windows?
D.windows
#      WindowGroup  ScanNumBegin  ScanNumEnd  ...  CollisionEnergy      mz_left     mz_right
# 0              1             0          36  ...             10.0  1265.384615  1301.384615
# 1              1            37          73  ...             10.0  1220.512821  1256.512821
# 2              1            74         110  ...             10.0  1175.641026  1211.641026
# 3              1           111         146  ...             10.0  1130.769231  1166.769231
# 4              1           147         183  ...             10.0  1085.897436  1121.897436
# ..           ...           ...         ...  ...              ...          ...          ...
# 520           21           735         771  ...             10.0   607.948718   643.948718
# 521           21           772         807  ...             10.0   563.076923   599.076923
# 522           21           808         844  ...             10.0   518.205128   554.205128
# 523           21           845         881  ...             10.0   473.333333   509.333333
# 524           21           882         918  ...             10.0   428.461538   464.461538

# [525 rows x 8 columns]



# Interested which frame collects data from which window group?
D.frames_meta()
#           Id         Time Polarity  ScanMode  ...  IsolationWidth  CollisionEnergy      mz_left     mz_right
# 0          1     0.326492        +         9  ...             NaN              NaN          NaN          NaN
# 1          2     0.434706        +         9  ...            36.0             10.0  1265.384615  1301.384615
# 2          3     0.540987        +         9  ...            36.0             10.0  1220.512821  1256.512821
# 3          4     0.648887        +         9  ...            36.0             10.0  1175.641026  1211.641026
# 4          5     0.756660        +         9  ...            36.0             10.0  1130.769231  1166.769231
# ...      ...          ...      ...       ...  ...             ...              ...          ...          ...
# 11548  11549  1243.494142        +         9  ...            36.0             10.0  1460.512821  1496.512821
# 11549  11550  1243.599171        +         9  ...            36.0             10.0  1415.641026  1451.641026
# 11550  11551  1243.707291        +         9  ...            36.0             10.0  1370.769231  1406.769231
# 11551  11552  1243.811222        +         9  ...            36.0             10.0  1325.897436  1361.897436
# 11552  11553  1243.919337        +         9  ...            36.0             10.0  1281.025641  1317.025641

# [11553 rows x 26 columns]
# This function joins the proper tables for you, so you don't have to.
# That how good that function is.

Vaex support

TimsPy offers support for a HDF5 based format that can be used with vaex. You can prepare the hdf file either directly from the terminal, or using the API. The command-line script is documented:

tims2hdf5.py --help

on Linux or macOS and

python tims2hdf5.py --help

on Windows results in

usage: tims2hdf5.py [-h] [--compression {gzip,lzf,szip,none}] [--compression_level COMPRESSION_LEVEL] [--chunksNo CHUNKSNO] [--shuffle]
                    [--columns {frame,scan,tof,intensity,mz,inv_ion_mobility,retention_time} [{frame,scan,tof,intensity,mz,inv_ion_mobility,retention_time} ...]] [--silent]
                    source target

Convert timsTOF Pro TDF into HDF5 with another set of parameters.

positional arguments:
  source                Path to the input .d folder containing TIMS data ("analysis.tdf" and "analysis.tdf_raw").
  target                Path to the output HDF5 file.

optional arguments:
  -h, --help            show this help message and exit
  --compression {gzip,lzf,szip,none}
                        Compression algorithm. (default: gzip)
  --compression_level COMPRESSION_LEVEL
                        Compression level, valid only for gzip compression. (default: 4)
  --chunksNo CHUNKSNO   Number of chunks. 0 for auto. (default: 0)
  --shuffle             Perform byte shuffling to help compression. (default: False)
  --columns {frame,scan,tof,intensity,mz,inv_ion_mobility,retention_time} [{frame,scan,tof,intensity,mz,inv_ion_mobility,retention_time} ...]
                        A list of columns to be included in the output HDF5 file. (default: frame scan tof intensity)
  --silent              Do not display progress bar. (default: False)

And the API can be used this way:

""" 
Here we show how to export data to veax-compatible format using API and how to read it in.

"""
from timspy.df import TimsPyDF
from timspy.vaex import TimsVaex

path = "/path/to/your/data.d"

D = TimsPyDF(path)
D.to_vaex("/path/to/where/to/save/data.hdf")
# above you can pass all other paramaters that are used by
# create_dataset function in h5py, as shown here:
# https://docs.h5py.org/_/downloads/en/2.6.0/pdf/

# Saving can take a while depending on the chosen algorithm.

V = TimsVaex("/path/to/where/to/save/data.hdf",
             "/path/to/your/data.d/analysis.tdf")# need sqlite

print(V)
TimsVaex.df
# #            frame    intensity    inv_ion_mobility    mz                  retention_time     scan    tof
# 0            1        9            1.6011418333261218  1174.6557905901582  0.326492080509831  33      312260
# 1            1        9            1.5999999999999996  733.4809407066116   0.326492080509831  34      220720
# 2            1        9            1.5999999999999996  916.952388791982    0.326492080509831  34      261438
# 3            1        9            1.5977164032508784  152.3556513940331   0.326492080509831  36      33072
# 4            1        9            1.5977164032508784  827.3114212681397   0.326492080509831  36      242110
# ...          ...      ...          ...                 ...                 ...                ...     ...
# 404,183,872  11553    9            0.6097471205799778  1171.1371127745708  1243.91933656337   909     311606
# 404,183,873  11553    9            0.6086254290341188  677.899262990993    1243.91933656337   910     207399
# 404,183,874  11553    9            0.6075037601757217  1084.4095190266144  1243.91933656337   911     295164
# 404,183,875  11553    9            0.6030173116029844  262.0435011990699   1243.91933656337   915     82016
# 404,183,876  11553    9            0.6018957561715719  1129.4302518267864  1243.91933656337   916     303778


# Then you can plot TIC 
V.plot_TIC(recalibrated=False)

To get a better understanding of what vaex can do, we refer you to (its manual)[https://vaex.readthedocs.io/en/latest/tutorial.html]. Below, let us show you how to use it to calculate the sum of observed intensities per scan.

import vaex
import matplotlib.pyplot as plt

I_per_scan = V.df.groupby(by='scan',
                          agg={'SummedIntenisty': vaex.agg.sum('intensity')})

# Then you can take the output, which is small and RAM-friendly,
# and do whatever you want with it :)
I_per_scan = I_per_scan.to_pandas_df().sort_values('scan')
#      scan  SummedIntenisty
# 629    33          2966589
# 630    34          2822660
# 631    35          2670091
# 868    36          2542816
# 632    37          2445243
# ..    ...              ...
# 624   913           372112
# 625   914           359317
# 626   915           365467
# 627   916           347171
# 628   917           347208

# [885 rows x 2 columns]

The calculation is done on multiple threads. Each thread gets its portion of the data automatically.

Plans

  • specialized methods for DDA experiments
  • going fully open-source with scan2inv_ion_mobility and tof2mz
  • reduction of size of the HDF5 files

Installation

pip install timspy

or for devel version:

pip install -e git+https://github.com/MatteoLacki/timspy/tree/devel

or with git:

git clone https://github.com/MatteoLacki/timspy
cd timspy
pip install -e .

To install vaex support, use

pip install timspy[vaex]

or install add in additional modules

pip install vaex-core vaex-hdf5

which seems much less of a hastle than figuring out how pypi internals work.

API documentation

Please visit our documentation page.

Too bloat?

We have a simpler module too. Check out our timsdata module.

Best wishes,

Matteo Lacki & Michal (Startrek) Startek