jungmannlab · rafalkowalewski1 · Mar 10, 2026 · Feb 24, 2026 · Feb 24, 2026 · Feb 24, 2026
diff --git a/.bumpversion.cfg b/.bumpversion.cfg
@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.9.7
+current_version = 0.9.8
 commit = True
 tag = False
 parse = (?P<major>\d+)\.(?P<minor>\d+)\.(?P<patch>\d+)(\-(?P<release>[a-z]+)(?P<build>\d+))?
@@ -15,10 +15,6 @@ serialize =
 
 [bumpversion:file:./picasso/version.py]
 
-[bumpversion:file:./release/pyinstaller/picasso.spec]
-
-[bumpversion:file:./release/pyinstaller/picassow.spec]
-
 [bumpversion:file:./release/one_click_windows_gui/picasso_innoinstaller.iss]
 
 [bumpversion:file:./release/one_click_windows_gui/create_installer_windows.bat]

diff --git a/.gitignore b/.gitignore
@@ -36,7 +36,9 @@ picasso/config.yaml
 
 # One-click-installer output
 release/one_click_windows_gui/Output/
+release/one_click_windows_gui/*.zip
 release/one_click_macos_gui/*.dmg
+release/one_click_macos_gui/*.zip
 
 # PyInstaller
 #  Usually these files are written by a python script from a template

diff --git a/changelog.rst b/changelog.rst
@@ -1,7 +1,38 @@
 Changelog
 =========
 
-Last change: 23-FEB-2026 CEST
+Last change: 10-MAR-2026 CEST
+
+0.9.8
+-----
+Small improvements:
++++++++++++++++++++
+- Added a function ``picasso.lib.get_save_filename_ext_dialog`` that can also check for the existence of the files with other extenstions (for example, if the user tries to save a .yaml file with the same name as an existing .hdf5 file, it will ask if the user wants to overwrite the .hdf5 file). This is implemented in all GUI modules when saving files.
+- ``PyImarisWriter`` is included in the one-click-installer again (Windows only)
+- Localize GUI allows the user to automatically undrift localizations
+- Localize Parameters dialog displays a message if the z calibration path in the config file could not be found
+- MLE fitting saves CRLB uncertainties of fitted parameters: photons, background, sx and sy
+- ``picasso.localize.fit`` default method changed to ``sigmaxy`` (anisotropic sigma fitting)
+- Render export localizations supports exporting all channels sequentially
+- Changed default max. frames in linking (dark times calculation) to 3 (previously 1) (both GUI and ``picasso.postprocess.link``)
+- Added number of binding events to Render's "Show info" dialog
+- Render 3D window always brings the selected region's mean z position to 0 for easier visualization
+- Render 3D: added buttons for xy, xz and yz projections
+- Added DOIs related to G5M and axial loc. precision
+- Removed mean frame filtering for G5M filtering/postprocessing
+- Added tool tips to G5M dialog
+- G5M automatically saves the check on relative sigma
+- Updated Picasso Average documentation
+- Changed default parameters in Simulate to reflect a typical DNA origami measurement
+- SPINNA GUI allows for user-defined max y-axis value in the NND plot
+
+Bug fixes:
+++++++++++
+- Fixed Picasso Server launching in one-click-installers
+- Fixed 3D MLE fitting and cleaned the docstrings for better readability (``picasso.gaussmle``)
+- Fixed how Picasso: Simulates splits photons across binding events
+- Fixed G5M 3D CI test
+- Fixed Render 3D scale bar
 
 0.9.7
 -----
@@ -82,8 +113,8 @@ Bug fixes:
 -------
 Important updates:
 ^^^^^^^^^^^^^^^^^^
-- **Algorithm for molecular mapping introduced (G5M)**, see documentation `here <https://picassosr.readthedocs.io/en/latest/render.html#g5m>`__ *DOI will be added once available*.
-- **Localize outputs axial localization precision for astigmatic imaging in 3D**,  *DOI will be added once available*. 
+- **Algorithm for molecular mapping introduced (G5M)**, see documentation `here <https://picassosr.readthedocs.io/en/latest/render.html#g5m>`__. DOI: `10.1038/s41467-026-70198-5 <https://doi.org/10.1038/s41467-026-70198-5>`_
+- **Localize outputs axial localization precision for astigmatic imaging in 3D**. DOI: `10.1038/s41467-026-70198-5 <https://doi.org/10.1038/s41467-026-70198-5>`_
 - Localize GUI allows the user to select which localization columns to save when saving localizations. See the new dialog in the *File* -> *Select columns to save*
 - Localize accepts frame bounds to analyze only a subset of frames
 - Config file accepts z calibration .yaml paths so that they can be automatically loaded when changing between cameras

diff --git a/distribution/picasso.iss b/distribution/picasso.iss
@@ -2,10 +2,10 @@
 AppName=Picasso
 AppPublisher=Jungmann Lab, Max Planck Institute of Biochemistry
 
-AppVersion=0.9.7
+AppVersion=0.9.8
 DefaultDirName={commonpf}\Picasso
 DefaultGroupName=Picasso
-OutputBaseFilename="Picasso-Windows-64bit-0.9.7"
+OutputBaseFilename="Picasso-Windows-64bit-0.9.8"
 ArchitecturesAllowed=x64
 ArchitecturesInstallIn64BitMode=x64
 

diff --git a/docs/average.rst b/docs/average.rst
@@ -13,8 +13,8 @@ The averaging module uses 2D cross-correlation to determine the rotational and t
 2. Save the picked localizations by selecting ``File`` > ``Save picked localizations``.
 3. Load the resulting file with picked localizations into ``Picasso: Average`` by selecting ``File`` > ``Open`` or dragging and dropping it into the window. It will start a parallel pool to increase computational speed.
 4. ``Picasso: Average`` will immediately perform a center-of-mass alignment of the picked structures and display an average image. Rotational and translational alignment will follow in the next steps.
-5. Select ``Process`` > ``Parameters`` and adjust the ``Oversampling`` parameter. We recommend choosing the highest number at which the average image still appears smooth. High oversampling values result in substantial computational time and can cause artifacts. Hence, it might be useful to first use low oversampling to generate a less-refined average image and then perform subsequent averaging steps with higher oversampling for optimized resolution.
-6. Adjust the number of average iterations in the ``Iterations`` field. In most cases, a value of 10 is more than sufficient. If you are unsure about the computational time of the process, choose one iteration as a starting point. More iterations can be added later by repeating the processing steps. After a certain number of iterations, the average image will converge, meaning that it will not change with more iterations. If you experience that one localization spot is overemphasized, try again with less oversampling and few iterations. The 2D cross-correlation is prone to lock in bright spots.
-7. Select ``Process`` > ``Average`` to perform particle averaging with the current oversampling for the set number of iterations. This step can be repeated with different settings. The program will use the current average image as a starting point.
+5. Select ``Process`` > ``Parameters`` and adjust the ``Display pixel size (nm)`` parameter. We recommend choosing the highest number at which the average image still appears smooth. Low display pixel size results in substantial computational time and can cause artifacts. Hence, it might be useful to first use high values to generate a less-refined average image and then perform subsequent averaging steps with lower values for optimized resolution.
+6. Adjust the number of average iterations in the ``Iterations`` field. In most cases, a value of 10 is more than sufficient. If you are unsure about the computational time of the process, choose one iteration as a starting point. More iterations can be added later by repeating the processing steps. After a certain number of iterations, the average image will converge, meaning that it will not change with more iterations. If you experience that one localization spot is overemphasized, try again with higher display pixel size and few iterations. The 2D cross-correlation is prone to lock in bright spots.
+7. Select ``Process`` > ``Average`` to perform particle averaging with the current display pixel size for the set number of iterations. This step can be repeated with different settings. The program will use the current average image as a starting point.
 8. Once the average image has converged, save the transformed localizations by selecting ``File`` > ``Save``. The resulting HDF5 localization file contains the aligned localizations in the center of the movie dimensions. It can be loaded like any other HDF5 localization file into ``Picasso: Render``.
 
diff --git a/docs/conf.py b/docs/conf.py
@@ -26,7 +26,7 @@
 # The short X.Y version
 version = ""
 # The full version, including alpha/beta/rc tags
-release = "0.9.7"
+release = "0.9.8"
 
 # -- General configuration ---------------------------------------------------
 
@@ -37,7 +37,9 @@
 # Add any Sphinx extension module names here, as strings. They can be
 # extensions coming with Sphinx (named 'sphinx.ext.*') or your custom
 # ones.
-extensions = []
+extensions = [
+    "sphinx.ext.mathjax",  # for rendering math equations
+]
 
 # Add any paths that contain templates here, relative to this directory.
 templates_path = ["_templates"]

diff --git a/docs/render.rst b/docs/render.rst
@@ -87,25 +87,25 @@ Upon clicking ``Perform RESI analysis``, each of the loaded channels is clustere
 G5M
 ---
 
-In Picasso 0.9.5, a new algorithm for molecular mapping (i.e., finding the positions of individual molecules from localizations) was introduced: G5M (Gaussian Mixture Modeling with Modifications for Molecular Mapping; Kowalewski, Reinhardt et al., *DOI will be added once available*). G5M is based on Gaussian Mixture Modeling (GMM) but includes several modifications to make it suitable for molecular mapping. All the technicalities as well as the user guide of the method are explained in the publication mentioned and its Supplementary Information. Please refer to ``picasso.g5m`` for the details of the implementation. Below is a brief summary of the user guide.
+In Picasso 0.9.5, a new algorithm for molecular mapping (i.e., finding the positions of individual molecules from localizations) was introduced: G5M (Gaussian Mixture Modeling with Modifications for Molecular Mapping; Kowalewski, Reinhardt et al. *Nature Comms*, 2026. DOI: 10.1038/s41467-026-70198-5). G5M is based on Gaussian Mixture Modeling (GMM) but includes several modifications to make it suitable for molecular mapping. All the technicalities as well as the user guide of the method are explained in the publication mentioned and its Supplementary Information. Please refer to ``picasso.g5m`` for the details of the implementation. Below is a brief summary of the user guide.
 
 G5M requires some preprocessing of localizations to filter out the badly fitted ones, especially the ones arising from crosstalk (overlapping blinking). These can be excluded from 2D data where the ellipticity and size of the image of an emitter in x and y can be filtered (in Picasso these are found under names “ellipticity”, “sx” and “sy”, respectively). Moreover, the photon count can be cut-off as crosstalk is likely to result in a higher-intensity signal. In 3D data these filters are less reliable due to astigmatism, however, “d_zcalib” could be used. We strongly encourage avoiding dense blinking, where emission signals from neighboring molecules overlap, especially during 3D image acquisition.
 
 Prior to molecular mapping, clustering of localizations is required to split the data into smaller chunks. For many datasets, DBSCAN works well. While in some cases some adjustments may be needed, we recommend the following DBSCAN parameters: In 2D, DBSCAN radius (epsilon) of 2*LP, in 3D - 3*LP (LP - average localization precision of the dataset, for example, NeNA or median localization precision). Default min. samples is set to 4. Clustering in Picasso adds the ``group`` column to the localization file, which is required for G5M. **Note: G5M relies on the information in the ``group`` column, therefore, if it is overwritten (for example, by picking localizations after DBSCAN clustering), G5M will not work.**
 
 To account for fluorophore non-specific sticking, frame analysis is normally recommended (especially the filtering of st. dev. of frame per molecule). However, if localizations from neighboring localization clouds overlap, this is not sufficient due to ambigous assignment of localizations to molecules. Therefore, we recommend filtering of molecules that express too few binding events (saved in the column ``n_events``). In the publication, we recommend a threshold of at least 3 binding events per molecule.
 
-The final postprocessing step is log-likelihood filtering (using the column ``p_val``). The recommended threshold is ``> 0.0015``.
+The final postprocessing step is log-likelihood filtering (using the column ``p_val``). The recommended threshold is ``> 0.0015``, however, it might need to be adjusted for your data, especially in 3D this can be too conservative.
 
-As a final check for overfitting (i.e., too many assigned molecules), G5M automatically saves a bar plot of the number of binding events per molecule (``n_events`` column) for sparse (with neighbors within 25 nm) and clustered (without neighbors within 80 nm) molecules. While this is only a qualitative verification, it is a simple method to spot potential overfitting issues. If the clustered molecules show fewer binding events that the sparse molecules, overfitting likely occurred. See Fig. S15 of the publication for an example of well-behaved data.
+As a final check for overfitting (i.e., too many assigned molecules), G5M automatically saves a bar plot of the number of binding events per molecule (``n_events`` column) for sparse (with neighbors within 25 nm) and clustered (without neighbors within 80 nm) molecules. While this is only a qualitative verification, it is a simple method to spot potential overfitting issues. If the clustered molecules show fewer binding events that the sparse molecules, overfitting likely occurred. See Fig. S15 of the publication for an example of well-behaved data. As of v0.9.8, Picasso saves the plot showing relative σ values (i.e., the fitted Gaussian σ divided by the average loc. precision around the molecule). This can be used to estimate if the loc. precision values are accurate (if not, many molecules will have relative σ values close to the user-selected min./max. σ).
 
 If the outcome of G5M seems unsatisfactory, please check the following:
 
 - Make sure that ``group`` column is present in the localization file and contains the correct information (i.e., from DBSCAN clustering, not from picking localizations);
 - Make sure that the loc. precision values (columns ``lpx``, ``lpy``, ``lpz``) are correct, comparing NeNA and median loc. precision is a reasonable proxy (without fiducial markers); the most common issue is a miscalibrated camera, leading to incorrect photon counts and thus incorrect loc. precisions;
 - Another reason why the loc. precision values can be off is due to the small box size in the localization step; especially in 3D astigmatic imaging, single-emitter images can be quite large, potentially exceeding the user-defined box size; in such cases, we recommend increasing the box size in the localization step and rerunning the analysis;
 - Inspect if the localizations were preprocessed as described above;
-- Rerun the analysis without postprocessing (filtering) and redo it manually, since the step may be too stringent, especially for short acquisition times;
+- Rerun the analysis without postprocessing (filtering) and redo it manually, since some steps may be too stringent, such as ``p_val`` or ``n_events`` (latter especially for short acquisition times);
 - Adjust min./max. σ, especially too low max. σ may lead to high false positive error rates (i.e., overfitting); We suggest inspecting ``rel_sigma`` values of the assigned molecules, which are calculated as the fitted σ divided by the mean localization precision of the surrounding localizations. If the values are close to the user-selected min./max. σ, min./max. σ might need to be adjusted. Alternatively, this might be a sign of inaccurate/inprecise loc. precision values, see above;
 - Adjust min. locs;
 - Adjust DBSCAN (or other clustering algorithm) parameters. For example, if G5M takes too long to run, the DBSCAN clusters most likely contain too many molecules. In such a case, we recommend splitting such clusters further;

diff --git a/docs/table01.csv b/docs/table01.csv
@@ -6,8 +6,8 @@ photons ,"The total number of detected photons from this event, not including ba
 sx ,The Point Spread Function width in camera pixels.,float 
 sy ,The Point Spread Function height in camera pixels.,float 
 bg ,"The number of background photons per pixel, not including the camera offset.",float 
-lpx ,"The localization precision in x direction, in camera pixels, as estimated by the Cramer-Rao Lower Bound of the Maximum Likelihood fit (Mortensen et al., Nat Meth, 2010). ",float 
-lpy ,"The localization precision in y direction, in camera pixels, as estimated by the Cramer-Rao Lower Bound of the Maximum Likelihood fit (Mortensen et al., Nat Meth, 2010). ",float 
+lpx ,"The localization precision in x direction, in camera pixels, as estimated by the Cramer-Rao Lower Bound (Mortensen et al., Nat Meth, 2010 and Smith et al., Nat Meth, 2010). ",float 
+lpy ,"The localization precision in y direction, in camera pixels, as estimated by the Cramer-Rao Lower Bound (Mortensen et al., Nat Meth, 2010 and Smith et al., Nat Meth, 2010). ",float 
 net_gradient ,"The net gradient of this spot which is defined by the sum of gradient vector magnitudes within the fitting box, projected to the spot center. ",float 
 z,(Optional) The z coordinate fitted in 3D in nm. Please note the units are different for x and y coordinates.,float 
 lpz,(Optional) The localization precision in z direction in nm.,float
@@ -20,4 +20,8 @@ len ,"(Optional) The length of the event in frames, if localizations from consec
 n ,"(Optional) The number of localizations in this event, if localizations from consecutive frames have been linked, potentially diverging from the “len” column due to a transient dark time tolerance.",long 
 photon_rate ,"(Optional) The mean number of photons per frame, if localizations from consecutive frames have been linked. The total number of photons is set in the “photons” column. ",float 
 x_pick_rot ,"(Optional) Projection of localizations onto the axis of the rectangular pick. Only available after saving rectangular pick(s). ",float
-y_pick_rot ,"(Optional) Projection of localizations against the axis of the rectangular pick. Only available after saving rectangular pick(s). ",float
+y_pick_rot ,"(Optional) Projection of localizations against the axis of the rectangular pick. Only available after saving rectangular pick(s). ",float
+photons_unc ,"(Optional) The uncertainty of the photons estimation as estimated by the Cramer-Rao Lower Bound of the Maximum Likelihood fit. ",float
+bg_unc ,"(Optional) The uncertainty of the background estimation as estimated by the Cramer-Rao Lower Bound of the Maximum Likelihood fit. ",float
+sx_unc ,"(Optional) The uncertainty of the sx estimation (camera pixels) as estimated by the Cramer-Rao Lower Bound of the Maximum Likelihood fit. ",float
+sy_unc ,"(Optional) The uncertainty of the sy estimation (camera pixels) as estimated by the Cramer-Rao Lower Bound of the Maximum Likelihood fit. ",float
diff --git a/picasso/__main__.py b/picasso/__main__.py
@@ -2402,7 +2402,7 @@ def main():
         "g5m",
         help=(
             "Gaussian Mixture Modeling with Modifications for Molecular Mapping"
-            "\nFor more details see: *DOI will be added once available*"
+            "\nFor more details see https://doi.org/10.1038/s41467-026-70198-5"
         ),
     )
     g5m_parser.add_argument(

diff --git a/picasso/clusterer.py b/picasso/clusterer.py
@@ -965,7 +965,8 @@ def test_subclustering(
     do not (sparse). If subclustering occurs, the clustered molecules
     should have a lower number of events on average.
 
-    DOI: *DOI will be added once available*
+    Introduced in Kowalewski, Reinhardt, et al. Nature Comms, 2026.
+    DOI: https://doi.org/10.1038/s41467-026-70198-5
 
     Parameters
     ----------