Changelog

v1.4.0 | 2024-02-28

New features

Bug fixes

  • Fixed a bug where the time indicator was off by one frame in ImageStack.plot_correlated() and ImageStack.export_video().

  • Fixed a bug where the time between frames was incorrectly not excluded when calling ImageStack.frame_timestamp_ranges() with include_dead_time=False. Note that Scan and Kymo are not affected.

  • Fixed a bug where ImageStack.plot_correlated() was not excluding the dead time between frames.

  • Fixed a bug where color adjustments on a single channel ImageStack would not be applied unless a channel was provided as argument. This bug was introduced in v1.3.0.

  • Changed the DateTime tag on TIFFs exported with Pylake from Scan and Kymo objects. Before the change, the start and end of the scanning period in nanoseconds was stored. After the change, we store the starting timestamp of the frame, followed by the starting timestamp of the next frame to be consistent with data exported from Bluelake. The scanning time is stored in the field Exposure time (ms) on the Description tag.

  • Fixed tests to be compatible with pytest>=8.0.0.

  • Ensure in returns True for a valid data path (e.g. "Force HF/Force 1x" in file should return True).

Improvements

  • ImageStack now closes any file handles it has open upon garbage collection.

Deprecations

  • Deprecated GaussianMixtureModel.from_channel(). The class constructor now accepts Slice instances directly.

  • Deprecated GaussianMixtureModel.label(). Use GaussianMixtureModel.state_path() instead. Note, the returned instance is type Slice rather than np.ndarray. The data can be accessed via model.state_path(channel_slice).data.

  • Deprecated GaussianMixtureModel.exit_flag Use GaussianMixtureModel.fit_info() instead. Note, the returned instance is a GmmFitInfo dataclass with attributes matching the keys of the dict returned from exit_flag (along with bic and aic; see next point).

  • Deprecated GaussianMixtureModel.bic and GaussianMixtureModel.aic properties. These values can now be accessed via the bic and aic properties of GaussianMixtureModel.fit_info.

Breaking changes (alpha functionality)

Other changes

  • Bump tifffile dependency to >=2022.7.28.

v1.3.1 | 2023-12-07

Bug fixes

v1.3.0 | 2023-12-07

New features

Bug fixes

  • Fixed a bug in Scan.plot() in which the default aspect ratio was calculated such that pixels always appeared square. For scans with non-square pixel sizes, this would result in distortion of the image.

  • Fixed a bug in Slice.downsampled_like that would fail to raise an error due to a lack of overlap between the low frequency and high frequency channel when the high frequency channel starts within the last sample of the low frequency channel.

  • Fixed lk.download_from_doi() to align with new Zenodo REST API.

  • Don’t store animation writer in a temporary variable as this results in a matplotlib error when attempting to export a movie on jupyter notebook.

Improvements

  • Kymographs consisting of a single scan line now return a valid line_time_seconds. This allows certain downstream functionality, such as Kymo.plot().

  • Issue a more descriptive error when attempting to compute a diffusion constant of a track with no points.

  • Pylake can now handle kymographs that were erroneously stored in the Scan field. Kymographs with a pre-specified number of lines to record were incorrectly being marked on the timeline and exported as ‘Scan’ instead of ‘Kymograph’ in versions of Bluelake prior to 2.5.0.

v1.2.1 | 2023-10-17

Bug fixes

  • Fixed lk.download_from_doi() to align with new Zenodo REST API.

  • Fixed a bug where the minimum length field of an exported KymoTrackGroup was formatted as an integer resulting in rounding errors when storing the tracks. Note that an incorrect minimum length can lead to biases when performing dwell time analysis. These values are now properly formatted as floating point numbers. The entry in the header was also changed to “minimum observable duration (seconds)” for additional clarity. This bug was introduced in version 1.2.0.

  • Fixed a bug that prevented resaving a KymoTrackGroup loaded from an older version of Pylake.

  • Fixed a bug that inadvertently made us rely on cachetools>=5.x. Older versions of cachetools did not pass the instance to the key function resulting in a TypeError: key() missing 1 required positional argument: '_' error when accessing cached properties or methods.

  • Fixed a bug that could cause a division by zero while fitting power spectra with f_diode. The lower bound of f_diode was changed from 0 to 1.0 Hz. This change should not impact results unless users were getting failed calibrations with this error.

  • Fixed a layout issue that made the sliders appear so narrow in jupyterlab that they were not visible.

v1.2.0 | 2023-08-15

New features

  • Added fitting mode "simultaneous" to lk.refine_tracks_gaussian() which enforces optimization bounds between the peak positions. This helps prevent lk.refine_tracks_gaussian() from reassigning points to the wrong track when a track momentarily disappears and overlap_strategy is set to "multiple" and refine_missing_frames is set to True. When fitting mode is set to "simultaneous", bounds ensure that the individual Gaussians cannot switch position. In addition, this mode uses a better initial guess for the peak amplitudes based on the maximum photon count observed in each range.

  • Added the option to take into account discretization effects in DwelltimeModel by passing a discretization_timestep to the model when constructing it.

  • Added the option to take into account discretization effects when performing dwell time analysis on a KymoTrackGroup. Simply pass discrete_model=True to KymoTrackGroup.fit_binding_times() to make use of this new functionality.

  • Added the optional parameter loss_function to fit_power_spectrum(). Implemented loss functions are "gaussian" (default) and "lorentzian". The default corresponds to regular least-squares fitting, whereas "lorentzian" invokes a robust fitting method that is less susceptible to spurious peaks in the power spectrum which comes at the cost of a small bias in the estimates for a spectrum without noise peaks. Furthermore, no estimates of the errors in the fitted parameters are provided. This is beta functionality. While usable, this has not yet been tested in a large number of different scenarios. The API can still be subject to change without any prior deprecation notice!

  • Added PowerSpectrum.identify_peaks() method to the PowerSpectrum class. This method uses probability to identify peaks in the spectrum that are not due to the movement of beads in an optical trap. This is beta functionality. While usable, this has not yet been tested in a large number of different scenarios. The API can still be subject to change without any prior deprecation notice!

  • Added KymoTrack.plot_fit() and KymoTrackGroup.plot_fit() to show the fitted model obtained from gaussian refinement.

  • Added the ability to specify a cropping region when exporting to an h5-file using file.save_as(filename, crop_time_range=(starting_timestamp, ending_timestamp)).

  • Added method to create colormaps approximating a color from emission wavelength. See lk.colormaps.from_wavelength() for more information.

  • Added support for accessing Kymo, Scan and PointScan by path (e.g. file["Kymograph"]["my_kymo"] or file["Kymograph/my_kymo"]).

  • Added support for slicing PointScan.

Bug fixes

  • Fixed issue in dwell time analysis that could lead to biased estimates for kymographs with very few events. Before this fix KymoTrackGroup.fit_binding_times() relied on the assumption that the shortest observed track is actually the minimum observable dwell time. This assumption is valid for kymographs with many events; however, this becomes problematic when multiple kymographs with few events each are analyzed globally. In this case, binding times will be underestimated. With this fix, the minimum observable dwell time is calculated from the kymograph scan line time and the set minimum track length.

    • The old (incorrect) behavior is maintained as default until the next major release (v2.0.0) to ensure backward compatibility.

    • To enable the fixed behavior immediately (recommended), specify observed_minimum=False when calling KymoTrackGroup.fit_binding_times().

    • To maintain the old legacy behavior use observed_minimum=True.

    • Note that CSVs exported from the kymotracking widget before v1.2.0 will contain insufficient metadata to make use of the improved analysis. To recover this metadata, use lk.filter_tracks() on the KymoTrackGroup with a specified min_length. This filters short events and stores the new minimum observable duration in the group.

Other changes

  • File.save_as() data now allows passing in a single string for the omit_data parameter.

  • Gracefully handle empty Scan after slicing. Previously, a slice operation on a Scan that resulted in no frames remaining raised a NotImplementedError. Now it returns an EmptyScan.

  • Improved performance of Scan.pixel_time_seconds, Kymo.pixel_time_seconds and Kymo.line_time_seconds.

  • Dropped opencv dependency which was only used for calculating rotation matrices and performing the affine transformations required for image alignment. Pylake now uses scikit-image for this purpose.

  • Marked functions that take file paths as arguments with the os.PathLike type hint to idicate that pathlib.Path and similar types are also accepted (not just str).

  • Use Unicode characters for µ and ² when plotting rather than TeX strings.

Deprecations

  • Deprecated fitting mode "multiple" in lk.refine_tracks_gaussian() as it could lead to spurious track crossings. See the entry for the fitting mode "simultaneous" under New Features for more information.

v1.1.1 | 2023-06-13

  • Fixed parameter description for minimum length. It now correctly reads that increasing this parameter reduces tracking noise.

  • Rarely used but heavy packages like opencv, scikit-image, and scikit-learn are no longer loaded eagerly with import lumicks.pylake. Rather, they are loaded on-demand the first time that a feature needs them. This makes importing pylake itself faster and mitigates potential third-party issues (e.g. this makes the first issue from the F.A.Q. less severe as it no longer affects all users).

v1.1.0 | 2023-05-17

New features

  • Enabled adding KymoTrackGroup instances with tracks from different source kymographs. Note that certain features (e.g., KymoTrackGroup.ensemble_msd() and KymoTrackGroup.ensemble_diffusion("ols")) require that all source kymographs have the same pixel size and scan line times. See the kymotracking tutorial for more information.

  • Added DwelltimeModel.profile_likelihood() for calculating profile likelihoods for dwell time analysis. These can be used to determine confidence intervals and assess whether the model has too many parameters (i.e., if a model with fewer components would be sufficient). See the population dynamics tutorial for more information.

  • Added functionality to split tracks in the Kymotracking widget.

  • Added functionality to provide minimum and maximum observation times per data point to DwelltimeModel and DwelltimeBootstrap.

Bug fixes

  • Fixed a bug that would round the center of the window to the incorrect pixel when using KymoTrack.sample_from_image(). Pixels are defined with their origin at the center of the pixel, whereas this function incorrectly assumed that the origin was on the edge of the pixel. The effects of this bug are larger for small windows. For backward compatibility this bugfix is not enabled by default. To enable the bugfix specify correct_origin=True as a keyword argument to the method.

  • Fixed bug that could lead to mutability of Kymo and Scan through .get_image() or .timestamps. This bug was introduced in 0.7.2. Grabbing a single color image from a Scan or Kymo using .get_image() returns a reference to an internal image cache. This numpy array was write-able and modifying data in the returned image would result in future calls to .get_image() returning the modified data rather than the original Kymo or Scan image. Timestamps obtained from .timestamps were similarly affected. These arrays are now returned non-writeable.

  • Fixed a bug where we passed a SampleFormat tag while saving to tiff using tifffile. However, tifffile infers the SampleFormat automatically from the datatype of the numpy array it is passed. This produced warnings on the latest version of tifffile when running the benchmark.

  • Ensure that the probability density returns zero outside its support for DwelltimeModel.pdf(). Prior to this change, non-zero values would be returned outside the observation limits of the dwelltime model.

  • Fixed a bug where DwelltimeModel.hist() produces one less bin than requested with n_bins.

  • Fixed a bug where lk.dsdna_ewlc_odijk_distance() erroneously returned a model where force is the dependent variable rather than distance. This bug was introduced in 0.13.2. Note that lk.ewlc_odijk_distance() was unaffected. It was specifically the convenience function with DNA parameters that was affected.

Improvements

  • Improve API description of the channel data in a default FdCurve.

  • Improved fitting performance dwell time analyses by implementing analytic gradient.

  • Added lower and upper bounds when fitting dwell time models. This prevents numerical issues that occur when approaching these extremes. Prior to this change, over-parameterized models could encounter numerical issues due to divisions by (near) zero. After this change, amplitudes are constrained to stay between 1e-9 and 1.0 - 1e-9, while lifetimes are constrained to remain between 1e-8 and 1e8.

  • Provided better initial guess for dwell time models with three or more components. Prior to this change, the dwelltime models involving three components or more would get an initial guess with an average lifetime beyond the mean of the observations.

  • Silenced false positive warnings about the optimization violating the bound constraints. Prior to this change, dwell time analysis would produce warnings about the optimizer exceeding bounds. These excursions beyond the bound are very small and have no effect on the model.

  • Parameters corresponding to Gaussian refinement are now preserved when two refined tracks are concatenated.

  • Set the minimum value for the Min length slider in the Kymotracking widget to 2.

v1.0.0 | 2023-03-14

Happy π-day!

New features

  • Added links to example data for all tutorials. Try running them in a jupyter notebook!

  • Added API to create a Kymo from an ImageStack. See ImageStack.to_kymo() for more information.

  • Added API for accessing notes in a h5 file, i.e. file.notes returns a dictionary of notes.

  • Implemented bias correction for centroid refinement that shrinks the window to reduce estimation bias. This bias correction can be toggled by passing a bias_correction argument to lk.track_greedy() and lk.refine_tracks_centroid(). Note that this bias correction is enabled by default.

  • Implemented spatial calibration for ImageStack. Images are now plotted in units of microns for TIF files which contain pixel size in the metadata. If the metadata is unavailable, the axes units are provided in pixels.

  • Added option to show scale bars to image plots using lk.ScaleBar(). See plotting and exporting for an example.

  • Added KymoTrackGroup.remove() to remove a KymoTrack from a KymoTrackGroup.

  • Enabled boolean array indexing (e.g. tracks[[False, False, True]]) and indexing with arrays of indices (e.g. tracks[[1, 3]]) for KymoTrackGroup. See the API documentation for more information.

  • Improved force calibration documentation. See the new Tutorial and Theory section.

  • Added API to look up (pixel)size of an ImageStack. See ImageStack for more information.

  • Enabled forwarding plot style arguments to Slice.range_selector(). Prior to this change, a plot with markers was always opened which resulted in very poor performance when plotting high frequency data.

Breaking changes

  • Pylake now requires Python version >= v3.9 and matplotlib version >= v3.5

  • When performing particle tracking on kymographs, bias correction is now enabled by default; without this correction, kymographs with high background signal will suffer from biased localization estimates. To disable bias correction, specify bias_correction=False to lk.track_greedy() and lk.refine_tracks_centroid().

  • The function ImageStack.define_tether() now takes coordinates in calibrated spatial units (microns) rather than pixels. If calibration to microns is not available, units are still in pixels.

  • Optional arguments for lk.track_greedy() and lk.track_lines() are now enforced as keyword only.

  • The vel argument has been renamed to velocity for lk.track_greedy().

  • lk.track_lines() now performs bias-corrected centroid refinement after tracking to improve localization accuracy. Note that the old behaviour can be recovered by passing refine=False.

  • When removing tracks with the kymotracking widget, only tracks that are entirely in the selection rectangle will be removed. Prior to this change, any tracks intersecting with the selection rectangle would be removed.

  • Added checks which enforce KymoTracks in a KymoTrackGroup to be unique. Extending a KymoTrackGroup with KymoTrack instances that are already part of the group will now result in a ValueError. Similarly, constructing a new KymoTrackGroup with duplicate KymoTracks will also produce a ValueError.

  • Removed all deprecated methods and properties from Kymo, Scan, PointScan, and ImageStack.

  • Removed deprecated methods KymoTrack.estimate_diffusion_ols() and KymoTrackGroup.remove_lines_in_rect()

  • Removed deprecated functions filter_lines(), refine_lines_centroid() and refine_lines_gaussian().

  • Removed deprecated property lines and method save_lines from KymoWidgetGreedy.

  • Removed deprecated argument exclude from Kymo.line_timestamp_ranges() and Scan.frame_timestamp_ranges(). Use the argument include_dead_time instead.

  • Removed deprecated argument line_width from lk.track_greedy().

  • Changed several asserts to Exceptions.

    • Attempting to read a KymoTrackGroup from a CSV file that doesn’t have the expected file format will result in an IOError.

    • Attempting to extend a KymoTrackGroup by a KymoTrackGroup tracked on a different Kymo now results in a ValueError.

    • Attempting to connect two tracks with the same start and ending time point now raises a ValueError.

    • FdFitter: FdFit.fit() now raises a RuntimeError when a fit has no data or fittable parameters.

    • FdFitter: FdFit.plot() now raises a KeyError when trying to plot data that does not exist.

    • FdFitter: FdFit.plot() now raises a RuntimeError when trying to plot a fit with multiple models without selecting a model using angular brackets [] first.

    • FdFitter: FdFit.profile_likelihood() now raises a ValueError when max_step <= min_step or max_chi2_step <= min_chi2_step.

    • FdFitter: when inverting a model with interpolate=True, Pylake now raises a ValueError if the minimum or maximum is not finite.

    • FdFitter: a ValueError is raised when adding incompatible models.

    • FdFitter: When adding data to a fit, adding data with an unequal number of points for the dependent and independent variable will now raise a ValueError.

    • FdFitter: When adding data to a fit, adding data with more than one dimension will raise a ValueError.

    • FdFitting: Attempting to evaluate a parameter trace with lk.parameter_trace() for a parameter that is not part of the model now results in a ValueError.

    • FdFitting: Attempting to compute a parameter trace while providing an incomplete set of parameters will now result in a ValueError.

    • Attempting to use Slice.downsampled_over() or Slice.downsampled_like() with timestamps ranges or another channel that doesn’t overlap with the channel now produces a ValueError.

    • Attempting to construct a TimeSeries where the length of the timestamp array is not equal to the length of the data array results in a ValueError.

    • FdEnsemble alignment now produces a ValueError if fewer than 2 datasets are provided.

    • Image reconstruction now raises a ValueError if the length of the data and infowave are not equal.

    • Plotting: When creating a plot providing axes and an image_handle, a ValueError is raised when those axes do not belong to the image_handle provided.

    • Widefield: Attempting to open multiple TIFF as a single ImageStack will now raise a ValueError if the alignment matrices of the individual TIFF are different.

    • PowerSpectrum: Attempting to replace the power spectral values of a PowerSpectrum using with_spectrum using a vector of incorrect length will raise a ValueError.

Bug fixes

  • Fixed incorrect behaviour in lk.track_lines() by interpolating back to integer frame times. Prior to this change, lk.track_lines() would provide a subpixel accurate position along the time axis of the kymograph as well. However, this position was specified with respect to the coordinate system of the image, rather than actual acquisition times. As such, it would produce incorrect results when performing downstream analysis that rely on the time corresponding to an actual time. Note that lk.track_greedy() is not affected.

  • Fixed bug in lk.track_lines() where one extra line was returned rather than the number requested through the parameter max_lines.

  • Fixed regression that caused ImageStack to incorrectly store the stack name (resulting in an extra (' in front of the name in plots). This regression was introduced in v0.13.0 when enabling multi-file support. See ImageStack for more information.

v0.13.3 | 2023-01-26

New features

  • Added KymoTrackGroup.ensemble_diffusion() for estimating an average diffusion constant for a collection of tracks.

  • Added diffusion theory section which details methods used for quantifying diffusive motion. Note that this documentation can be downloaded and run inside a Jupyter notebook.

  • Added KymoTrack.plot() and KymoTrackGroup.plot() methods to conveniently plot the coordinates of tracked lines. See the kymotracking documentation for more details.

  • Added the lk.GaussianMixtureModel.extract_dwell_times() method to calculate dwell-times for all states from channel data.

  • Added lk.colormaps with custom colormaps for plotting single-channel images. Note: this is a namedtuple so you can access the attributes using the dot notation (for example kymo.plot("blue", cmap=lk.colormaps.cyan)). The available attributes are red, green, blue, magenta, yellow, cyan.

  • Added localization_variance to DiffusionEstimate. This quantity is useful for determining the diffusive SNR.

  • Added variance_of_localization_variance to DiffusionEstimate when calculating an ensemble CVE. This provides an estimate of the variance of the averaged localization uncertainty.

  • Added option to pass a localization variance and its uncertainty to KymoTrack.estimate_diffusion().

  • Added option to calculate a diffusion estimate based on the ensemble MSDs using the Ordinary Least Squares (OLS) method.

  • Added DwelltimeBootstrap.extend() to draw additional samples for the distribution.

Bug fixes

  • Fixed a bug that resulted in an incorrect round-off of the window size to pixels when kymotracking. This bug resulted in using one more pixel on each side than intended for specific track_widths. Track width is selected by rounding to the next odd window size. Prior to this change, the number of points used would increase on even window sizes. As a result, requesting a track width of 2.5 pixels, would result in using a window of size 5. Currently, requested a track width of 3 pixels results in a window of size 3, while 3.1 rounds up to the next odd window size (5). This bug affected the kymo tracking widget (tracking, refinement and photon count sampling during saving), lk.track_greedy() and lk.refine_tracks_centroid().

  • Updated default slider ranges for the Kymotracker widget to reflect minimum track width required for tracking.

  • Fixed issue with model description not being available in Jupyter notebooks for some force-distance models.

  • Show validity criterion for Marko Siggia WLC models in terms of model parameters. Prior to this change the limit was simply denoted as 10 pN where in reality it depends on the model parameters. The 10 pN was a reasonable value for most DNA constructs.

  • Fixed bug which occurred when exporting images to TIFF files of a numerical type with insufficient range for the data with the flag clip=True. Prior to this change, values exceeding the range of the numerical type were not clearly defined. After this change values below and above the supported range are clipped to the lowest or highest value of the data type respectively.

  • Fixed bug in DwelltimeBootstrap.hist() (previously named DwelltimeBootstrap.plot(), see below). Previously, only up to two components were plotted; now all components are plotted appropriately.

  • DwelltimeBootstrap.hist() now displays the original parameter estimate rather than the mean of the bootstrap distribution; the bootstrap distribution is used solely to calculate the confidence intervals via DwelltimeBootstrap.get_interval().

  • Fixed a bug where Scan.export_video() and ImageStack.export_video() would show elements from a previous plot.

  • Fixed a bug that caused a misalignment of half a pixel between the kymograph and its position histogram when using Kymo.plot_with_position_histogram().

Deprecations

  • Renamed CorrelatedStack to ImageStack.

  • Deprecated the DwelltimeModel.bootstrap attribute; this attribute will be removed in a future release. Instead, DwelltimeModel.calculate_bootstrap() now returns a DwelltimeBootstrap instance directly. See the population dynamics documentation for more details.

  • Deprecated DwelltimeBootstrap.plot() and renamed to DwelltimeBootstrap.hist() to more closely match the figure that is generated.

  • Deprecated DwelltimeBootstrap.calculate_stats(). This method is replaced with DwelltimeBootstrap.get_interval() which returns the 100*(1-alpha) % confidence interval; unlike DwelltimeBootstrap.calculate_stats(), it does not return the mean.

Other changes

  • DwelltimeBootstrap is now a frozen dataclass.

  • Attempting to access DwelltimeModel.bootstrap before sampling now raises a RuntimeError; however, see the deprecation note above for proper API to access bootstrapping distributions.

  • Suppress legend entry for outline when invoking KymoTrack.plot().

  • Allow pickling force calibration results (CalibrationResults). Prior to this change two functions involved in calculating upper parameter bounds prevented this class from being pickled.

v0.13.2 | 2022-11-15

New features

  • Added covariance-based estimator (cve) option to KymoTrack.estimate_diffusion(). See kymotracker documentation for more details.

  • TIFFs exported from Scan and Kymo now contain metadata. The DateTime tag indicates the start/stop timestamp of each frame. The ImageDescription tag contains additional information about the confocal acquisition parameters.

  • Added the Kymo.duration property to provide convenient access to the total scan time in seconds.

  • Added addition operator to KymoTrackGroup. KymoTrackGroups tracked on the same Kymo can be concatenated with the + operator.

  • Added the optional min_length parameter to KymoTrackGroup.estimate_diffusion() to discard tracks shorter than a specified length from the analysis.

  • Added the DwelltimeModel.rate_constants property along with additional documentation explaining the assumptions that underlie using the exponential model. See the dwell time analysis documentation for more information.

  • Ensure same call signature for plot() methods for Scan, Kymo, PointScan and CorrelatedStack:

    • Scan, Kymo and PointScan: Made argument channel optional.

    • Scan and PointScan: Added argument show_title.

    • Kymo: Added arguments image_handle and show_title.

    • CorrelatedStack: See deprecation changelog entry.

  • Introduced lazy loading for TimeSeries data. This means that the data corresponding to a TimeSeries channel is not read from disk until it is used.

Bugfixes

  • Added a check which verifies that a Kymo is not downsampled prior to estimating a diffusion constant. Computing diffusion constants from temporally downsampled kymographs is now explicitly disallowed.

  • Fixed a bug in KymoTrack.estimate_diffusion() that could lead to biased estimates obtained with the "ols" estimator. For the diffusion estimate itself to be affected, specific lags have to be missing from the track (for example every second point in a track). This is a regression that was introduced in v0.12.1.

  • Added a warning to KymoTrack.estimate_diffusion() used with the "ols" method when points are missing from a track. In this case the uncertainty estimate is biased. See the section on diffusive processes for more details.

  • Added a warning that estimating the optimal number of points to use when using the "ols" method can be biased if many points are missing.

  • Fixed a bug in KymoTrack.estimate_diffusion() that could lead to biased estimates obtained with the "gls" estimator when gaps occur in the track. Such cases now produce an exception recommending the user to refine the track prior to diffusion estimation. See the section on diffusive processes for more details.

  • Fixed issue where on Jupyter Lab the kymotracker widget would align the Kymograph and track parameters vertically rather than horizontally.

  • Functions that use KymoTrackGroup now gracefully handle the cases where no tracks are available. The refinement functions refine_tracks_centroid() and refine_tracks_gaussian() return an empty list, while KymoTrackGroup.fit_binding_times() and KymoTrackGroup.plot_binding_histogram() raise an exception.

  • lk.track_greedy() now returns an empty KymoTrackGroup instead of an empty list when no coordinates exceed the threshold.

  • lk.track_greedy() now returns an empty KymoTrackGroup instead of an error when an ROI is selected that results in no lines tracked.

  • Fixed a bug where the pixel_threshold could be set to zero for an empty image. Now the minimum pixel_threshold is one.

  • Fixed a bug where single pixel detections in a KymoTrackGroup would contribute values with a dwell time of zero. These are now dropped, the correct minimally observable time is set appropriately and a warning is issued.

  • Fixed slicing of a Kymo where slicing from a time point inside the last line to the end (e.g. kymo["5s":]) resulted in a Kymo which returned errors upon trying to access its contents.

  • Fixed a minor bug in force calibration. In rare cases it was possible that the procedure to generate an initial guess for the power spectral fit failed. This seemed to occur when the spectrum supplied is a mostly flat plateau. After the fix, an alternative method to compute an initial guess is applied in cases where the regular method fails. Note that successful calibrations were not at risk for being incorrect due to this bug since they would have resulted in an exception rather than an invalid result.

Deprecations

  • Deprecated property CorrelatedStack.src. In future versions, the contents of src will be considered an implementation detail that is not directly accessible. Data should be accessed through the public API.

  • Reordered the keyword arguments of the method CorrelatedStack.plot() and enforced all parameters after channel to be keyword arguments. For details see the docstring.

  • Enforced the argument axes of the method plot() for Scan, Kymo and PointScan to be a keyword argument.

  • Renamed force distance fitting functions. They are deprecated now and will be removed in the future:

    • inverted_marko_siggia_simplified -> wlc_marko_siggia_distance

    • marko_siggia_simplified -> wlc_marko_siggia_force

    • marko_siggia_ewlc_distance -> ewlc_marko_siggia_distance

    • marko_siggia_ewlc_force -> ewlc_marko_siggia_force

    • odijk -> ewlc_odijk_distance

    • inverted_odijk -> ewlc_odijk_force

    • freely_jointed_chain -> efjc_distance

    • inverted_freely_jointed_chain -> efjc_force

    • twistable_wlc -> twlc_distance

    • inverted_twistable_wlc -> twlc_force

Other changes

  • All KymoTrack instances must have the same source Kymo and color channel in order to be in the same KymoTrackGroup instance. While this behavior was required previously for some downstream analyses on the tracks, it is now explicitly enforced upon KymoTrackGroup construction.

  • When calling KymoTrackGroup.estimate_diffusion() without specifying the min_length parameter, tracks which are shorter than the required length for the specified method will be discarded from analysis and a warning emitted. Previously, if any tracks were shorter than required, an error would be raised.

  • Updated benchmark to not use deprecated functions and arguments. Prior to this change, running the benchmark would produce deprecation warnings.

  • Kymo.plot() now returns a handle of the plotted image.

  • PointScan.plot() now returns a list of handles of the plotted lines.

v0.13.1 | 2022-09-08

Bug fixes

  • Reverted lazy loading for TimeSeries data as this actually caused significantly long wait times for accessing the data.

v0.13.0 | 2022-09-06

New features

  • Added possibility to access property sample_rate for TimeSeries data with constant sample rate.

  • Allow reading multiple files with lk.CorrelatedStack (e.g. lk.CorrelatedStack("image1.tiff", "image2.tiff")).

  • Added CorrelatedStack.export_video() to export videos to export multi-frame videos to video formats or GIFs.

  • Added support for steps when slicing frames from CorrelatedStacks.

  • Added function Kymo.line_timestamp_ranges() to obtain the start and stop timestamp of each scan line in a Kymo. Please refer to Confocal images for more information.

  • Added Kymo.flip() to flip a Kymograph along its positional axis.

  • Added KymoTrackGroup.estimate_diffusion() to estimate diffusion constants for a group of kymograph traces.

  • Include unit in DiffusionEstimate dataclass.

  • Added shape property to Scan and Kymo.

  • Allow slicing CorrelatedStacks with timestamps and time strings (e.g. stack["5s":"10s"] or stack[f.force1x.start:f.force1x.stop]).

  • Allow slicing Scan with timestamps and time strings (e.g. scan["5s":"10s"] or scan[f.force1x.start:f.force1x.stop]).

  • Allow downloading files directly from Zenodo using lk.download_from_doi(). See the example on Cas9 binding for an example of its use.

  • Made piezo tracking functionality public and added piezo tracking tutorial.

  • Lazily load data and timestamps for TimeSeries data

  • Propagate Slice axis labels when performing arithmetic (when possible).

  • Added a warning to the Kymotracker widget if the threshold parameter is set too low, which may result in slow tracking and the widget hanging.

  • Added header line to exported track coordinate CSV files. The first header line now contains the version of Pylake which generated the file and a version number for the CSV file itself (starting with v2 from this release).

  • It is now possible to pickle FdFit objects. Prior to this change, unpickling an FdFit would fail since model identification relied on a stored id for each of the models used. The id of a variable changes whenever a new variable is created however. After this change, each model is associated with a universally unique identifier (uuid) that is used for identification instead. This uuid is serialized with the Model and used by FdFit thereby preserving their relationship when pickled/unpickled.

Bug fixes

  • Improved scan.get_image("rgb") to handle missing channel data. Missing channels are now handled gracefully. Missing channels are zero filled matching the dimensions of the remaining channels.

  • Added calls to manually redraw the axes in the kymotracker widget during horizontal pan and line connection callbacks. Without this, the plot did not update correctly when using newer versions of ipywidgets and matplotlib.

  • Fixed a bug in the video export that led to one frame less being exported than intended.

  • Fixed a bug which prevented the range selector widget from updating when the dataset to be plotted is changed. Previously, on some supported versions of matplotlib it would no longer update the figure. This is now fixed.

  • Force distance models now raise a ValueError exception when simulated for invalid parameter values.

  • Force distance models now have a non-zero lower bound for the contour length (Lc), persistence length (Lp), stretch modulus (St) and Boltzmann constant times temperature (kT) instead of a lower bound of zero.

  • Force distance fits now raise a RuntimeError if any of the returned simulation values are NaN.

  • Fixed a bug that resulted in profile likelihood automatically failing when an attempted step exceeded the bounds where the model could be simulated.

Breaking changes

  • To disable image alignment for lk.CorrelatedStack, the alignment argument has to be provided as a keyword argument (e.g. lk.CorrelatedStack("filename.tiff", align=False) rather than lk.CorrelatedStack("filename.tiff", False)).

  • Removed deprecated argument roi from CorrelatedStack.export_tiff. Use CorrelatedStack.crop_by_pixels() to select the ROI before exporting.

  • CorrelatedStack.frame_timestamp_ranges() is now a method rather than a property. This was done for API consistency with the API for Scan. Please refer to Correlated stacks for more information.

  • Removed public attributes CorrelatedStack.start_idx and CorrelatedStack.stop_idx and made them protected.

  • The property sample_rate of Continuous data now returns a float instead of an int.

  • Changed the error type when attempting to access undefined per-pixel timestamps in Kymo from AttributeError to NotImplementedError.

  • KymoWidgetGreedy now enforces using keywords for all arguments after the first two (kymo and channel).

  • The following KymoWidgetGreedy attributes/functions have been removed (replaced with private API): adding, algorithm_parameters, axis_aspect_ratio, output_filename, plotted_lines, show_lines, create_algorithm_sliders(), refine(), show(), track_all(), track_kymo() and update_lines().

  • The track_width argument of refine_tracks_centroid() expects values in physical units whereas the deprecated refine_lines_centroid() expected the line_width argument in pixel units.

  • Removed default value provided for driving_frequency_guess in lk.calibrate_force().

  • It is now mandatory to supply a sample_rate when calling lk.calibrate_force().

  • It is now mandatory to supply a sample_rate when calling lumicks.pylake.force_calibration.touchdown.touchdown().

Deprecations

  • Scan.save_tiff() and Kymo.save_tiff() were deprecated and replaced with Scan.export_tiff() and Kymo.export_tiff() to more clearly communicate that the data is exported to a different format.

  • Deprecated export_video_red(), export_video_green(), export_video_blue(), and export_video_rgb() methods for Scan. These methods have been replaced with a single export_video(channel=color) method.

  • In the functions Scan.frame_timestamp_ranges() and Kymo.line_timestamp_ranges(), the parameter exclude was deprecated in favor of include_dead_time for clarity.

  • Deprecated KymoTrackGroup.remove_lines_in_rect(); use KymoTrackGroup.remove_tracks_in_rect() instead (see below).

  • Deprecated the line_width argument of track_greedy(); use track_width instead.

  • Deprecated filter_lines(); use filter_tracks() instead.

  • Deprecated refine_lines_centroid(); use refine_tracks_centroid() instead. Note: the track_width argument of refine_tracks_centroid() expects values in physical units whereas the previous refine_lines_centroid() expected the line_width argument in pixel units.

  • Deprecated refine_lines_gaussian(); use refine_tracks_gaussian() instead.

  • Deprecated KymoWidget.save_lines(); use KymoWidget.save_tracks() instead.

Other changes

  • Added default values for the track_greedy() arguments track_width and pixel_threshold.

  • Renamed classes/methods/functions dealing with tracked particles. This change was made to avoid ambiguity with regard to the term “line”. Now, a “line” refers to a single scan pass of the confocal mirror during imaging. A “track” refers to the coordinates of tracked particles from a kymograph. Note: Any breaking changes or deprecations to the public API are noted above. The renamed classes/functions below are considered internal API and subject to change without notice; these classes should not be constructed manually:

    • KymoLine was renamed to KymoTrack

    • KymoLineGroup was renamed to KymoTrackGroup

    • export_kymolinegroup_to_csv() was renamed to export_kymotrackgroup_to_csv()

    • import_kymolinegroup_from_csv() was renamed to import_kymotrackgroup_from_csv()

  • Updated the KymoWidgetGreedy UI to reflect changes in terminology.

  • Made ipywidgets>=7.0.0 and notebook>=4.4.1 optional dependencies for pip.

  • Made notebook>=4.4.1 a mandatory dependency for conda release.

  • Pylake now depends on h5py>=3.4, <4. This change is required to support lazy loading for TimeSeries.

v0.12.1 | 2022-06-21

New features

  • Added Scan.pixel_time_seconds and Kymo.pixel_time_seconds to obtain the pixel dwell time. See Confocal images for more information.

  • Allow cropping CorrelatedStack using multidimensional indexing, i.e. stack[start_frame : end_frame, start_row : end_row, start_column : end_column]. See Correlated stacks for more information.

  • Added KymoLine.estimate_diffusion() which provides additional information regarding the diffusion constant estimation. This dataclass can be converted to floating point to get just the estimate, but also contains the number of points used to compute the estimate, and the number of lags used in the analysis. In addition to that, it provides a std_err field which reports the standard error for the estimate.

  • Added generalized least squares method as method option for KymoLine.estimate_diffusion(). Please refer to the kymotracker documentation for more information.

  • Added offline piezo tracking functionality (documentation pending).

  • Added lk.benchmark() that can be used to estimate the performance of your computer with various pylake tasks.

Bug fixes

  • Changed the internal calculation of the extent argument in Kymo.plot() such that the spatial limits are now defined at the center of the pixel (same functionality that is used for the temporal axis). Previously the limits were defined at the edge of the pixel resulting in a subtle misalignment between the coordinates of the tracked lines and the actual image.

  • Fixed an issue where the pixel dwell time stored in TIFFs exported from Pylake could be incorrect. Prior to this fix, Pylake exported the value entered in the Bluelake UI as pixel dwell time. In reality, the true pixel dwell time is a multiple of the acquisition sample rate. After the fix, TIFF files correctly report the achieved pixel dwell time.

  • Changed the internal calculation of the hydrodynamically correct force spectrum. Before this change, the computation of the power spectral density relied on the frequencies always being positive. While this change does not affect any results, it allows evaluating the power spectral density for negative frequencies.

  • Fixed an issue where evaluating the hydrodynamically correct spectrum up to very high frequencies could lead to precision losses. These precision losses typically occur at frequencies upwards of 30 kHz and manifest themselves as visible discontinuities.

  • Perform better input validation on the kymotracking functions track_greedy and track_lines. The line width and pixel threshold must be larger than zero. The diffusion parameter must be positive. Previously, failure to provide values respecting these limits would produce cryptic error messages.

  • Perform better input validation on refine_lines_centroid. Line width must now be at least one pixel. Previously, negative values produced a cryptic error message, while a line width smaller than one pixel would silently result in no refinement taking place.

  • Fixed bug in force calibration convenience function where setting fixed_alpha or fixed_diode to zero resulted in those parameters still being fitted. After this change, setting fixed_alpha to zero will result in the diode model having a fixed alpha of zero, whereas setting f_diode to zero raises an exception.

  • Include one extra sample when requesting frame timestamp ranges from a scan (Scan.frame_timestamp_ranges(exclude=True)). Previously, when slicing using these timestamps, you would omit the last sample of the scan. Now this sample will be included.

Deprecations

  • KymoLine.estimate_diffusion_ols() is now deprecated; use KymoLine.estimate_diffusion(method="ols") instead. This new method returns the same estimate of the diffusion coefficient as before but includes additional information about the fit.

v0.12.0 | 2022-04-21

New features

  • Support negating channels (e.g. neg_force = - file.force1x). See files and channels for more information.

  • Allow applying color intensity adjustments on images using lk.ColorAdjustment(). See Confocal images and Correlated stacks for more information.

  • Added DwelltimeModel to fit dwelltimes to an exponential (mixture) distribution. For more information, see the tutorials section on Population Dynamics

  • Allow slicing directly with an object with a .start and .stop property. See files and channels for more information.

  • Allow boolean array indexing on Slice (e.g. file.force1x[file.force1x.data > 5]. See files and channels for more information.

  • When performing arithmetic on Slice, the calibration is propagated if it is still valid.

  • Allow applying a gamma adjustment on images using lk.ColorAdjustment(). See Confocal images and Correlated stacks for more information.

  • Added lk.dsdna_odijk() and lk.ssdna_fjc() convenience functions to build Fd models for dsDNA and ssDNA with parameter defaults based on literature.

Bug fixes

  • Fixed a minor bug in KymoLineGroup.fit_binding_times(). Previously, the binding time for all lines in the group were used for the analysis. However, lines which start in the first frame of the kymo or end in the last frame have ambiguous dwelltimes as the start or end of the line is not known definitively. Now, the default behavior is to exclude these lines from the analysis. This behavior can be overridden with the keyword argument exclude_ambiguous_dwells=False. In general, this bug would lead to only very minor biases in the results unless the number of dwells to be excluded is large relative to the total number.

  • Fixed bug in vmax handling for CorrelatedStack. Before vmax values were scaled to the maximally possible range of values for the image instead of the actual intensity value. Note that use of vmax and vmin is deprecated and one should use adjustment=lk.ColorAdjustment(min, max) for color adjustments. See Correlated stacks for more information.

  • Fixed a bug in the kymotracker in which the plotted aspect ratio did not match the requested axis_aspect_ratio argument.

Deprecations

  • Deprecated red_image, green_image, blue_image, and rgb_image properties for Scan and Kymo. These data should now be accessed using the get_image(channel="{color}") method (where "{color}" can be "red", "green", "blue", or "rgb").

Breaking changes

  • Changed the frame indexing convention for plotting confocal scans to match CorrelatedStack.plot(). Previously, Scan.plot(frame=1) referred to the first frame in the stack. Now, indexing starts at 0.

  • Requesting a frame outside of the available range for Scan.plot() now throws an IndexError.

  • Removed deprecated properties scan_width_um, json, has_force, has_fluorescence from confocal classes.

Other changes

  • Changed titles for all plots of Scan and CorrelatedStack images to be consistent. First frame is titled as "[frame 1 / N]" and last frame is titled as "[frame N / N]".

  • The returned type from KymoLineGroup.fit_binding_times() has been changed to DwelltimeModel. Note, this class has the same public attributes and methods as the previously returned BindingDwelltimes class; however the plot() method has been deprecated and renamed to DwelltimeModel.hist(). This new method name more closely describes the actual functionality and also unifies the API with GaussianMixtureModel.

  • KymoLine is now immutable.

  • Removed examples directory. Application examples can be found in the online documentation.

v0.11.1 | 2022-02-22

New features

  • Added support for Python 3.10.

  • Added CorrelatedStack.define_tether() which can be used to define the endpoints of the tether between two beads and return image data rotated such that the tether is horizontal. Check out the documentation for more information.

  • Added function to correlate Scan frames to channel data. See Confocal images.

  • Added CorrelatedStack.crop_and_rotate() for interactive editing of the image stack. Actions include scrolling through image frames with the mouse wheel, and left-clicking to define the tether coordinates, and right-click/drag to define a cropping region. Check out the documentation for more information.

  • Added KymoLineGroup.plot_binding_histogram() to plot histograms of binding events for tracked lines. See Kymotracking for more details.

  • Allow fixing the photon background parameter in refine_lines_gaussian(). See kymotracking for more information.

  • Added option to specify a custom label when plotting fit with FdFit.plot(). See the tutorial section on Fd Fitting for more information.

  • Added functionality to slice Scan objects by frame indices. See Confocal images.

  • Added convenience function which allows users to perform a force calibration procedure with a single function call calibrate_force(). See force calibration.

  • Added Kymo.crop_and_calibrate() for interactive cropping of a kymograph. If the optional tether_length_kbp argument is supplied, the resulting kymograph will be automatically calibrated to this length. Check out the documentation for more information.

  • Added fallback to the function Kymo.plot_with_force() when only low-frequency force data is available.

  • Added option to undo/redo actions in the kymotracker widget.

  • Added option to fit peaks simultaneously in refine_lines_gaussian() using the flag overlap_strategy="multiple". See kymotracking for more information.

  • Added ability to manually connect two lines from any points (not just the ends) in the kymotracker widget.

Bug fixes

  • Fixed issue which resulted in the offset parameter added by Model.subtract_independent_offset() not having a unit associated with it.

  • Fixed bug which resulted in erroneous standard errors on parameter estimates computed from an FdFit with fixed parameters. For such a fitting problem, the covariance matrix was evaluated for the unconstrained problem (without the fixed parameter constraints). As a result, standard errors were always overestimated. Note that uncertainty estimation by profile likelihood was unaffected.

  • Fixed issue which resulted in overly stringent positional tolerance when using the kymotracker widget. This tolerance has now been made proportional to the axis viewport.

  • Removed axial parameter from lk.ActiveCalibrationModel() as we do not support active force calibration in the axial direction.

  • Improved default scaling behaviour for CorrelatedStack.plot_correlated() and Scan.plot_correlated(). It now ensures the ratio between the image and temporal plot is according to the aspect ratio of the scan or stack.

  • Slicing CorrelatedStack in reverse (i.e., stack[5:3]) or resulting in an empty stack (i.e., stack[5:5]) now throws an exception.

  • Resolved DeprecationWarning for tifffile.imsave() and tifffile.TiffWriter.save() with tifffile >= 2020.9.30.

  • Fixed a bug in refine_lines_gaussian() which incorrectly rounded the pixel position instead of flooring it. For pixels with a subpixel position larger than half the pixel, this resulted in shifting the window by one pixel in the positive direction. Typically, this would have little or no effect since the majority of the peak should still be covered.

Deprecations

  • Deprecated CorrelatedStack.timestamps and replaced with CorrelatedStack.frame_timestamp_ranges. The reason for this change is that per-pixel timestamps are not defined for camera based images; therefore, this previous use was not in line with the use of the timestamps property of confocal image classes. This change also brings consistency with Scan.frame_timestamp_ranges.

  • Deprecated plot_red(), plot_green(), plot_blue(), and plot_rgb() methods for PointScan, Scan, and Kymo. These methods have been replaced with a single .plot(channel={color}) method.

v0.11.0 | 2021-12-07

New force calibration features

  • Added support for active force calibration. For more information, please read: active force calibration.

  • Added support for axial force calibration. See axial calibration for more information.

  • Added support for using near-surface corrections for lateral and axial force calibration, please read: force calibration.

  • Added function to compute viscosity of water at specific temperature. Please refer to: force calibration for more information.

  • Added parameters describing the inferred driving peak (driving_amplitude, driving_frequency, driving_power) when performing active force calibration to CalibrationResults.

Other new features

  • Added Kymo.calibrate_to_kbp() for calibrating the position axis of a kymograph from microns to kilobase-pairs. Note: this calibration is applied to the full kymograph, so one should crop to the bead edges with Kymo.crop_by_distance() before calling this method.

  • Added CorrelatedStack.get_image() to get the image stack data as an np.ndarray.

  • Allow setting custom slider ranges for the algorithm parameters in the kymotracker widget. Please refer to kymotracker widget for more information.

  • Added function Scan.frame_timestamp_ranges() to obtain the start and stop timestamp of each frame in a Scan. Please refer to Confocal images for more information.

Bug fixes

  • Fixed issue in force calibration where the analytical fit would sometimes fail when the corner frequency is below the lower fitting bound. What would happen is that the analytical fit resulted in a negative term of which the square root was taken to obtain the corner frequency. Now this case is gracefully handled by setting the initial guess halfway between the lowest frequency in the power spectrum and zero.

  • Fixed issue that led to DeprecationWarning in the kymotracker widget.

  • Fixed error in kymotracking documentation. The tutorial previously indicated the incorrect number of samples for sample_from_image.

  • Fixed bug that would implicitly convert Kymograph and Scan timestamps to floating point values. Converting them to floating point values leads to a loss of precision. For more information see files and channels for more information.

  • Fixed bug where color-aligned data was returned from TiffFrame.data although alignment was not requested (e.g., CorrelatedStack("filename.tiff", align=False)). This bug was introduced in v0.10.1.

Deprecations

  • CorrelatedStack.raw has been deprecated and will be removed in a future release. Use CorrelatedStack.get_image() instead.

Breaking changes

  • Changed default for viscosity in force calibration models. When omitted, pylake will use the viscosity of water calculated from the temperature. Note that this results in the default (when no viscosity or temperature is set) changing from 1.002e-3 to 1.00157e-3 Pa*s. Please refer to force calibration for more information.

  • Units are now included in the headers for kymograph traces exported from the widget or with KymoLineGroup.save() (either um or kbp depending on the calibration of the kymograph). Any code that hardcoded the header names directly should be updated.

  • Added more input validation for model parameters when performing force calibration. We now force bead and sample density to be more than 100 kg/m³ when specified. Temperature should be specified between 5 and 90 °C and viscosity should be bigger than 0.0003 Pa*s (viscosity of water at 90 degrees Celsius).

v0.10.1 | 2021-10-27

New features

  • Added support for axial force detection (i.e. force detection along the Z axis). The high-frequency data can be accessed with f.force1z and f.force2z while the downsampled low-frequency channels can be accessed with f.downsampled_force1z and f.downsampled_force2z. The calibrations can be accessed with f.force1z.calibration and f.force2z.calibration. The Z component is not factored in the calculation of the total force f.downsampled_force1 and f.downsampled_force2.

  • Added KymoLineGroup.fit_binding_times() to allow for dwelltime analysis of bound states found using kymotracker. See kymotracking for more information.

  • Allow using data acquired with a fast force sensor by passing fast_sensor=True when creating a PassiveCalibrationModel.

  • Allow using hydrodynamically correct power spectrum when performing force calibration. See force calibration for more information.

  • Added ability to crop a CorrelatedStack. See Correlated stacks for more information.

Bug fixes

  • Show an error message when user attempts to refine lines before tracking or loading them so the kymotracker widget does not become unresponsive.

  • Force calibration models now throw an error when a bead diameter of less than 10^-2 microns is used (rather than produce NaN results).

  • Fixed bug that prevented export of CorrelatedStack if the alignment matrices were missing from the metadata.

  • Fixed bug where the exported metadata fields were erroneously exported as "Applied channel X alignment" for CorrelatedStack where the alignment was not actually applied.

  • Fixed error in optimal point determination for MSD estimator. Note that this bug led to an error being thrown when the number of points in the trace was exactly 4. Now it results in a more informative error message (You need at least 5 time points to estimate the number of points to include in the fit.).

Improvements

  • Switched to trust region reflective algorithm for fitting thermal calibration spectrum. This results in fewer optimization failures.

  • Pylake now ensures that f_diode stays below the Nyquist frequency during fitting.

  • Implemented a bias correction for the thermal calibration. Note that this typically leads to a small correction unless you use a very low number of points per block.

Deprecations

  • CorrelatedStack.from_data() has been renamed to CorrelatedStack.from_dataset() for consistency with BaseScan.from_dataset().

v0.10.0 | 2021-08-20

Important notice: This release contains an important fix that could lead to timestamp corruption when slicing kymographs. If you are still on version 0.8.2 or 0.9.0 we highly recommend updating or not using the kymograph slicing functionality (e.g. using the syntax kymo["0s":"5s"]). Please refer to the Bug fixes section for more information.

New features

  • Added option to exclude ranges with potential noise peaks from the calibration routines. Please refer to force calibration for more information.

  • Added crop_by_distance to Kymo to allow cropping Kymographs by distance. Please refer to kymographs.

  • Added refine_lines_gaussian() for refining lines detected by the kymotracking algorithm using gaussian localization. See kymotracking for more information.

Bug fixes

  • Fixed bug in Kymo which resulted in erroneous timestamps and line times after slicing the Kymograph by time (e.g. Kymo["1s":"5s"]). The reason for this was imprecision in the timestamp reconstruction that occurred when slicing the data. This in turn led to an erroneous reconstruction of the timestamps. As a result, any downstream analysis that relies on the time axis of a Kymograph post-slicing cannot be trusted (MSD analysis, plotting, tracking, attributes such as .line_time_seconds, etc.) in pylake v0.8.2 and v0.9.0. Only those two versions are affected. Note that regular image reconstruction was not affected. Timestamps of unsliced kymographs constructed directly from the .h5 files are also not affected.

  • Fixed an error in the documentation. In v0.9.0 PowerSpectrum.power was changed to represent power in V^2/Hz instead of 0.5 V^2/Hz. However, the docs were not appropriately updated to reflect this change in the model equation that’s fitted to the spectrum. This is mitigated now.

  • Fixed bug in plot_with_force which caused an exception on Kymographs with a partial last pixel.

Breaking changes

  • Removed the option to specify a custom reduce_timestamps function in kymo.downsampled_by. The reason for the removal is that by downsampling repeatedly with different reduce functions for the timestamps, the data can end up in an inconsistent state. Additionally, the timestamps of the original object (read directly from the h5 file) are defined as the mean of the pixel timestamps; this definition is now consistent regardless of downsampling state.

v0.9.0 | 2021-07-29

Pylake v0.9 provides several new features. Starting from pylake v0.9, the kymotracker will handle units for you. From now on, all you have to worry about is physical quantities rather than pixels. Please see the updated example on Cas9 binding for a demonstration of this. In addition to that, you can now infer diffusion constants from diffusive kymograph traces.

For convenience, we added the option to simulate force-distance models without having to fit them first and directly do arithmetic with channels. Pylake also supports baseline corrected force data from Bluelake now.

Unfortunately, some of these improvements required some breaking changes so please see the detailed changelog entries for more information.

New features

  • Added Kymo.downsampled_by() for downsampling Kymographs in space and time. See kymographs for more information.

  • Added option to stitch Kymograph lines visually via the Jupyter notebook widget.

  • Added Mean Square Displacement (MSD) and diffusion constant estimation to KymoLine. For more information, please refer to kymotracking

  • Added FdCurve.with_baseline_corrected_x() to return a baseline corrected version of the FD curve if the corrected data is available. Note: currently the baseline is only calculated for the x-component of the force channel in Bluelake. Therefore baseline corrected FdCurve instances use only the x-component of the force channel, unlike default FdCurves which use the full magnitude of the force channel by default.

  • Added ability to perform arithmetic on Slice (e.g. (f.force1x - f.force2x) / 2). For more information see files and channels for more information.

  • Allow simulating force model with a custom set of parameters (see the tutorial section on Fd Fitting for more information).

  • Added a slider to set the algorithm parameter velocity to the kymotracker widget.

Bug fixes

  • Fixed bug in kymotracker which could result in sample_from_image returning erroneous values in some cases. When sampling the image, we sample pixels in a region around the traced line. This sampling procedure is constrained to stay within the image bounds. Previously, we used the incorrect axis to clamp the pixel index along the positional axis. As a consequence, kymographs that are wider (in terms of number of pixels on the positional axis) than they are long (in terms of number of pixels along the time axis) would only have an accurate sampling in the top half of the kymograph. The lower portion of the kymograph would result in zero counts.

  • Fixed bug which could lead to bias when tracking lines using track_greedy() with the rectangle tool. Selecting which region to track used to pass that region specifically to the tracking algorithm. This means that the algorithm is unable to sample the image outside of the passed region. Therefore, this can lead to a bias when the line to be tracked is near the edge of the selected region. In the updated version, all image processing steps that depend on the image use the full image.

  • Fixed bug which could lead to bias when tracking lines using track_lines() with the rectangle tool. Selecting which region to track used to pass that region specifically to the tracking algorithm. This means that the blurring steps involved in this algorithm become biased near the edges (since they do not get contributions from outside the selected areas, while they should). In the updated version, all image processing steps that depend on the image use the full image.

  • Fixed a bug in Kymo.plot_with_force() which resulted in the plotting function throwing an error for Kymographs with an incomplete final line.

  • Fixed a bug in the plotting order of CalibrationResults.plot(). Previously, when plotting after performing a force calibration, the model fit was erroneously plotted first (while the legend indicated that the model fit was plotted last). The results of the calibration itself are unchanged.

  • Resolved DeprecationWarning with tifffile >= 2021.7.2.

  • Fixed a bug in CalibrationResults.ps_model_fit which resulted in its attribute num_points_per_block to be 1 rather than the number of points per block the model was fitted to. Note that this does not affect the calibration results as the calibration procedure internally used the correct number of points per block.

Breaking changes

  • Changed PowerSpectrum.power to actually reflect power in V^2/Hz. Before it was expressed in 0.5 V^2/Hz.

  • Dropped support for Python 3.6.

  • Pylake now depends on numpy>=1.20. This change is required to use a different fft normalization in the force calibration tests.

  • The attribute image_data in KymoLine is now private.

  • Make kymotracker functions track_greedy(), track_lines(), and class KymoWidgetGreedy take Kymo and a channel (e.g. “red”) as their input. The advantage of this is that now units of time (seconds) and space (microns) are propagated automatically to the tracked KymoLines. See the Cas9 kymotracking example or the kymotracking tutorial for more information.

  • KymoLineGroup.save() and KymoWidgetGreedy.save_lines() no longer take dx and dt arguments. Instead, the correct time and position calibration is now passed automatically to these functions. See kymographs for more information.

  • Express kymotracker algorithm parameters line_width, sigma, velocity and diffusion in physical units rather than pixels. Prior to this change, the units of the kymotracking algorithm were in pixels. Note that if you want to reproduce your earlier results multiply line_width and sigma by kymo.pixelsize_um[0], velocity by kymo.pixelsize_um[0] / kymo.line_time_seconds and diffusion by kymo.pixelsize_um[0] ** 2 / kymo.line_time_seconds.

  • In the FD Fitter, Parameters.keys() is now a member function instead of a property (used to be invoked as parameter.keys) to be consistent with dictionary.

  • Slice.downsampled_like() now returns both the downsampled Slice and a copy of the low frequency reference Slice cropped such that both instances have exactly the same timestamps. The reason for this is that the first two samples of the low frequency trace can typically not be reconstructed (since there is no high frequency data for those available). This led to confusion, since now the trace downsampled_like produces is shorter than the input. By returning both, this problem is mitigated. Please refer to Files and Channels for an example of its updated use.

  • Optimization settings are now passed to fit_power_spectrum() as keyword arguments instead of using the class lk.CalibrationSettings.

  • Renamed CalibrationResults.ps_model_fit and CalibrationResults.ps_fitted to CalibrationResults.ps_model and CalibrationResults.ps_data for clarity.

  • Drop units from parameter names in CalibrationResults. Note that the unit is still available in the .unit attribute of a calibration parameter.

  • Accessing an element from a force calibration performed with Pylake now returns a CalibrationParameter instead of a float. The calibration value can be accessed as calibration["kappa"].value, while the unit can be accessed in the .unit property.

v0.8.2 | 2021-04-30

New features

  • Added GaussianMixtureModel to fit a Gaussian Mixture Model to channel data. For more information, see the tutorials section on Population Dynamics

  • Added Scan.size_um and Kymo.size_um to return the scanned size along each dimension. Use these properties to access the scan sizes along each axis for scans and kymographs.

  • Added force calibration functionality to Pylake. Please refer to force calibration for more information.

  • Added support for new metadata format in order to handle widefield TIFF files recorded in Bluelake 1.7.

Bug fixes

  • Fixed bug in kymotracker which could result in a line being extended one pixel too far in either direction. Reason for the bug was that a blurring step in the peak-finding routine was being applied on both axes, while it should have only been applied to one axis. Note that while this bug affects peak detection (finding one too many), it should not affect peak localization as that is performed in a separate step.

  • Fixed bug in kymotracker slicing which could result in one line too many or too few being included. The bug was caused by using the timestamp corresponding to one sample beyond the last pixel of the line as “start of the next line” without accounting for the dead time that may be there.

Deprecations

  • Scan.scan_width_um and Kymo.scan_width_um have been deprecated. Use Scan.size_um and Kymo.size_um to get the scan sizes. When performing a scan, Bluelake determines an appropriate scan width based on the desired pixel size and the scan width entered in the UI. The property scan_width_um returned the scan size entered by the user, rather than the actual scan size achieved. To obtain the achieved scan size, use Scan.size_um and Kymo.size_um.

v0.8.1 | 2021-02-17

New features

  • Added widget to graphically slice FdCurve by distance in Jupyter Notebooks. It can be opened by calling pylake.FdDistanceRangeSelector(fdcurves). For more information, see the tutorials section on notebook widgets.

  • Added FdCurve.range_selector() and FdCurve.distance_range_selector(). See Notebook widgets for more information.

  • Added center_point_um property to PointScan, Kymo and Scan classes. Use these properties to access the metadata for scans and kymographs instead of the deprecated json field.

  • Added scan_width_um property to Kymo and Scan classes. Use these properties to access the metadata for scans and kymographs. instead of the deprecated json field.

  • Added FdCurve.with_offset() to FdCurve to add offsets to force and distance.

  • Added FdEnsemble to be able to process multiple FdCurve instances simultaneously. See FD Curves for more information.

  • Added FdEnsemble.align_linear() to align F,d curves in an ensemble by correcting for a constant offset in force and distance using two linear regressions. See FD curves for more information.

  • Added CorrelatedStack.export_tiff() for exporting aligned image stacks. See Correlated stacks for more information.

Bug fixes

  • Fixed bug when using continuous channels which lead to excessive memory usage and degraded performance.

  • Fixed Slice.downsampled_over() to ignore gaps rather than result in an unhandled exception. Previously when you downsampled a TimeSeries channel which had a gap in its data, Slice.downsampled_over() would try to compute the mean of an empty subsection, which raises an exception. Now this case is gracefully handled.

Breaking changes

  • Deprecated json attribute in confocal classes PointScan, Scan, and Kymo. Note: The format of the raw metadata exported from Bluelake is likely to change in future releases and therefore should not be accessed directly. Instead, use the accessor properties, as documented for scans and kymographs.

  • Slice.range_selector() is now a method instead of a property.

  • Deprecated has_force and has_fluorescence properties in confocal classes PointScan, Scan, and Kymo.

  • Renamed fd_selector.py to range_selector.py.

  • FdRangeSelectorWidget is no longer public.

  • Renamed FDCurve and FDSlice to FdCurve and FdSlice.

Other

v0.7.2 | 2020-01-14

New features

  • Support Python 3.9 (this required bumping the h5py requirement to >= 3.0).

  • Added refine_lines_centroid() for refining lines detected by the kymotracking algorithm. See kymotracking for more information.

  • Added Kymo.plot_with_force() for plotting a kymograph and corresponding force channel downsampled to the same time ranges of the scan lines.

  • Added Kymo.plot_with_position_histogram() and Kymo.plot_with_time_histogram() for plotting a kymograph and corresponding histogram along selected axis.

  • Added KymoLineGroup.save() for saving tracked Kymograph traces to a file. See kymotracking for more information.

  • Allow nbAgg backend to be used for interactive plots in Jupyter notebooks (i.e. %matplotlib notebook). Note that this backend doesn’t work for JupyterLab (please see the FAQ for more information).

  • Show downsampled_force channels when printing a File.

Bug fixes

  • Fixed exception message in Slice.downsampled_to() which erroneously suggested to use force=True when downsampling a variable frequency channel, while the correct argument is method="force".

  • Fixed bug in Kymo plot functions which incorrectly set the time limits. Now, pixel centers are aligned with the mean time for each line.

Other

  • Include ipywidgets as a dependency so users don’t have to install it themselves.

  • Switch opencv dependency to headless version.

v0.7.1 | 2020-11-19

  • Add start and stop property to Slice. These return the timestamp in nanoseconds.

  • Add start argument to Slice.plot() which allows you to use a specific timestamp as time point zero in a plot. See files and channels for more information.

  • Added Slice.downsampled_like to downsample a high frequency channel according to the timestamps of a low frequency channel, using the same downsampling method as Bluelake.

  • Fixed bug in Kymo plotting functions. Previously, the time limits were calculated including the fly-in/out times which could lead to subtle discrepancies when comparing against force channels. These dead times are now omitted.

  • Add workaround for Scan and Kymo which could prevent valid scans and kymos from being opened when the start timestamp of a scan or kymo had a timestamp earlier than the actual start of the timeline channels. The cause of this time difference was the lack of quantization of a delay when acquiring STED images. This delay resulted in a subsample offset between the Scan/Kymo start time and the start of the timeline data which was falsely detected as a corrupted Scan or Kymo.

v0.7.0 | 2020-11-04

New features

  • Added seconds attribute to Slice. See files and channels for more information.

  • Added downsampled_to to Slice for downsampling channel data to a new sampling frequency. See section on downsampling in files and channels.

  • Added save_as to File for exporting compressed HDF5 files or omitting specific channels from an HDF5 file. See tutorial on files and channels for more information.

  • Added Scan.export_video_red, Scan.export_video_green, Scan.export_video_blue and Scan.export_video_rgb to export multi-frame videos to video formats or GIFs. See tutorial on confocal images for more information.

  • Added widget to graphically slice a Slice in Jupyter Notebooks. It can be opened by calling channel.range_selector. For more information, see notebook widgets.

  • Added profile likelihood method to FdFitter. See confidence intervals and standard errors for more information.

  • Added support for RGB images to CorrelatedStack.

  • Added image alignment to CorrelatedStack. Image alignment is enabled by default and will automatically align the color channels for multi-channel images, provided that the alignment metadata from Bluelake is available. Image alignment can be disabled by specifying align=False when loading the CorrelatedStack. For more information see correlated stacks.

  • Exposed low-level kymograph API (still in alpha status!). See tutorial on kymotracking.

  • Added Kymo.line_time_seconds for obtaining the time step between two consecutive kymograph lines.

  • Added Kymo.pixelsize_um and Scan.pixelsize_um for obtaining the pixel size for various axes.

Bug fixes

  • Improved accuracy of covariance matrix computation. To compute the covariance matrix of the parameter estimates, it is required to estimate the standard deviation of the residuals. This calculation was previously biased by not correctly taking into account the number of degrees of freedom. This is fixed now.

  • Fixed bug in CorrelatedStack.plot_correlated which lead to the start index of the frame being added twice when the movie had been sliced.

  • Fixed bug in File.scans so that a warning is generated when a scan is missing metadata. Other scans that can be loaded properly are still accessible.

  • Fixed bug in plot_correlated which did not allow plotting camera images correlated to channel data when the channel data did not completely cover the camera image stack.

Breaking changes

  • Fixed bug in downsampled_over which affects people who used it with where="center". With these settings the function returns timestamps at the center of the ranges being downsampled over. In the previous version, this timestamp included the end timestamp (i.e. t1 <= t <= t2), now it has been changed to exclude the end timestamp (i.e. t1 <= t < t2) making it consistent with downsampled_by for integer downsampling rates.

  • Fixed bug in File.point_scans to return PointScan instances. Previously, attempts to access this property would cause an error due to missing PointScan.from_dataset method. Note that the __init__ method arguments of PointScan have been changed to be in line with Kymo and Scan.

v0.6.2 | 2020-09-21

  • Support plotting Z-axis scans. Z-axis scans would previously throw an exception due to how the physical dimensions were fetched. This issue is now resolved.

  • Add slicing (by time) for FdCurve.

  • Add widget to graphically slice FdCurve in Jupyter Notebooks. It can be opened by calling pylake.FdRangeSelector(fdcurves). For more information, see the tutorials section on notebook widgets.

  • Fixed bug in FdRangeSelectorWidget that prevented drawing to the correct axes when other axes has focus.

  • Fixed displayed coordinates to correctly reflect position in Kymo.plot_red(), Kymo.plot_green(), Kymo.plot_blue() and Kymo.plot_rgb(). The data origin (e.g. kymo.red_image[0, 0]) is displayed on the top left of the image in these plots, whereas previously this was not reflected correctly in the coordinates displayed on the plot axes (placing the coordinate origin at the bottom left).

v0.6.1 | 2020-08-31

  • Added inverted simplified Marko Siggia model with only entropic contributions to FdFitter.

  • Change exception that was being raised on non-API field access such as Calibration, Marker, FD Curve, Kymograph and Scan to a FutureWarning rather than an exception.

v0.6.0 | 2020-08-18

  • Plot and return images and timestamps for scans using physical coordinate system rather than fast and slow scanning axis. In v5.0, this resulted in a failed reconstruction and Scan.pixels_per_line being defined as pixels along the x axis. Scan.pixels_per_line and Kymo.pixels_per_line now return the number of pixels along the fast axis. This fix was also applied to the timestamps. In the previous version, for a multi-frame scan where the y-axis is the fast axis, you could incorrectly get the time between pixels on the fast axis by calculating scan.timestamps[0, 0, 1] - scan.timestamps[0, 0, 0]. In the new version, this is scan.timestamps[0, 1, 0] - scan.timestamps[0, 0, 0] (note that the image array convention is [frame, height, width]). Note that this is a breaking change affecting scans with the fast axis in y direction!

  • Verify starting timestamp when reconstructing Kymo or Scan. In those cases, scans cannot be reliably reconstructed from the exported data and an error is thrown. For kymos, the first (partial) line is omitted and a warning is issued. Note that scans where the scan was initiated before the exported time window cannot be reconstructed! For kymos, the first line cannot be reconstructed if the export window does not cover the start of the kymograph.

  • Add literature page to the documentation.

  • Fix docstring for Fit.plot().

  • Optimized reconstruction algorithm for sum.

v0.5.0 | 2020-06-08

  • Added F, d Fitting functionality (beta, see docs tutorial section Fd Fitting and examples Twistable Worm-Like-Chain Fitting and RecA Fd Fitting).

  • Fixed an issue which prevented images from being reconstructed when a debugger is attached. Problem resided in reconstruct_image which threw an exception when attempting to resize a numpy array while the debugger was holding a reference to it.

  • Fixed bug that lead to timestamps becoming floating point values when using channel.downsampled_over.

v0.4.1 | 2020-03-23

  • Drop matplotlib < 3 requirement.

  • Add functionality which redirects users to the API when accessing particular fields, e.g. accessing file["FD curve"] will throw an error and redirect users to use file.fdcurves.

  • Add API for markers, i.e. file.markers returns a dictionary of markers (see docs tutorials section: Files and Channels).

  • Bugfix CorrelatedStack.plot() which resulted in the function throwing an error rather than showing a single frame.

  • Add canvas draw call to CorrelatedStack.plot_correlated() to make sure the plot is also interactive when it is not run from an interactive notebook.

v0.4.0 | 2020-01-21

  • Add calibration data as attribute of force channels (see docs tutorials section: Files and Channels).

  • Fixed bug which produced shifted timestamps when slicing continuous channel with time values between two data points.

  • Export pixel size and dwell time to exported TIFFs.

  • Implement slicing for kymographs (see docs tutorials section: Kymographs).

  • Show keys in a group when printing the group.

  • Allow iteration over groups.

  • Fix bug which could cause axes to be inverted for reconstructed images.

  • Add functionality to correlate images recorded from cameras with timeline data (see docs tutorial section: Correlated stacks).

v0.3.1 | 2019-03-27

  • Loading scans and kymographs is now much faster

  • Improved perfomance of slicing continuous channels

  • Fixed Unknown channel kind error with the v2 file format

  • Fixed deprecation warnings with h5py v2.9

v0.3.0 | 2018-12-04

  • TIFF scan exports now default to using 32bit floats

  • Support for photon time tag data has been added (Bluelake version 1.5.0 and higher, HDF5 file format version 2)

  • File properties like File.kymos, File.scans now return empty dict objects instead of an error if there are no such items in the file

v0.2.0 | 2018-07-27

  • Channel slices can be downsampled: lf_force = hf_force.downsampled_by(factor=20)

  • FdCurves now support subtraction, e.g. fd = f.fdcurves["measured"] - f.fdcurves["baseline"]

  • Scans and kymos now have a .timestamps property with per-pixel timestamps with the same shape as the image arrays

  • Added Matlab compatibility examples to the repository in examples/matlab

  • h5py >= v2.8 is now required

v0.1.0 | 2018-06-20

  • Initial release