Data correction and quality control¶
Chaining¶
A fluorophore that is on for multiple frames in the raw data will appear as a series of localizations at sequential times. To group localizations close in space and time into single events, run . A dialog will appear allowing chaining options to be set.
- class FindClumps
- Clump radius
is the maximum spatial distance between chained localizations. The default is twice the localization’s lateral fit error (a \(2\sigma\) should correctly link 95% of localisations).
- Time window
is the maximum temporal distance (in frames) allowed between chained localizations.
Pressing OK in the dialog will then identify which localizations are likely members of a chain, but will not replace the members of the chain with a single grouped/averaged localisation. This is done in a separate step, .
Drift correction¶
PYMEVisualise supports 3 forms of drift correction out of the box, with additional algorithms available as plugins. The builtin methods are as follows:
- Fiducial based drift correction
This uses fiducials localised along with the blinking molecules to correct drift, and assumes that the fiducial localisations are present in a different dataset to the single molecule localisations (as optimal detection, background subtraction, and fitting settings are likely to be different for fiducials and molecules). If the data was analysed using PYME, both these datasets should be in the same file, and running fiducial based correction should be as simple as selecting from the menu (and potentially entering the fiducial diameter to permit filters to be set accordingly). If fiducial and single molecule datasets are not in the same file, you will need to load the localisations first and then run to load the fiducial fits. The algorithm extracts fiducial traces, and aligns and averages the traces from multiple fiducials (weighted by localisation precision). It tolerates small gaps in the fiducial traces as long as not all fiducials traces are broken in the same frame. After correction is complete, , will show the residuals (error between each fiducial and the average correction) which gives an indication of correction quality.
- Autocorrelation based drift correction
Accessed as , this is essentially an implementation of the algorithm described in the supplement of [huang2008], dividing the localizations into overlapping time blocks, and performing autocorrelation between those blocks.
- Transmitted light correction
This relies on drift measurements made during image acquisition using a transmitted light channel (see [mcgorty2013] - our implementation does not assume, however, that correction of x-y drift is necessarily performed in real-time, rather saving the recorded drift values along with the image data), and requires suitable drift event data in the input files.
Fourier Ring Correlation¶
Fourier ring correlation (FRC) is an established technique for estimating the resolution of localization-based images
[nieuwenhuizen2013]. To use FRC to estimate resolution in PYMEVisualize, first select
. The will create 2 fake colour channels, block0 and block1,
based on dividing localisations in time with a temporal block size set using the dialog in Fig. 3 a (for
multicolour data it will split the existing color channels in 2 so you will get double the number of colour channels,
e.g. chan0_block0, chan0_block1, chan1_block0, chan1_block1).
Render images by choosing from the menu and selecting the two blocks corresponding to the color channel of interest (the FRC module currently assumes a single color), as shown in Fig. 3 b. Note that in principle any image generation method (see Image Reconstruction) can be used, but histogram rendering is probably the best for a pure resolution assessment.
In the rendered image window, choose and select the renderings of the two time blocks to compare as in Fig. 3 c. An FRC plot like Fig. 3 d will appear, quantifying resolution.
Fig. 3 Dialogs and plots in the Fourier ring correlation pipeline. (a) Dialog for , used to set FRC time block size. (b) Histogram generation dialog window. Pixel size is set to 5 nm
and FRC block0 and block1 are selected for rendering. (c) Dialog for ,
indicating blocks to compare for FRC. (d) FRC plot for image shown in Fig. 2 a.¶
Photophysics¶
For a given image, it is possible to estimate the photophysics of the dye or fluorescent protein used in acquisition. To do this, first run through the clump detection part of the chaining procedure described in Chaining. Then select . This will display three graphs, shown in Fig. 4, indicating the fluorescence decay rate of the fluorophore, the mean number of fluorophores in an ON state per second throughout the duration of imaging, and the mean number of photons per frame.
Note that the metadata setting Camera.CycleTime, which is the integration time of the camera used to collect the
raw localization data, must be present in order to analyze photophysics. See Metadata for details on how to
ensure this metadata is present.
Fig. 4 Plots generated from running on data shown in Ratiometric colour settings. (a) Estimation of fluorophore decay rate, indicated as \(\tau\) in the upper right of the plot. (b) Estimation of mean number of fluorophores in an ON state per second throughout the duration of imaging, indicated as \(\tau\) in the upper right of the plot. (c) Estimation of the mean mean number of photons per fluorophore in the ON state, indicated as Ph. mean in the upper right of the plot.¶
- nieuwenhuizen2013
Nieuwenhuizen et al., “Measuring image resolution in optical nanoscopy,” Nat. Methods, vol. 10, no. 6, pp. 557–562, 2013.
- huang2008
Huang, W. Wang, M. Bates, and X. Zhuang, “3D super-res imaging by STORM,” Science (80-. )., vol. 319, no. 5864, pp. 810–813, 2008.
- mcgorty2013
McGorty, D. Kamiyama, and B. Huang, “Active microscope stabilization in three dimensions using image correlation,” Opt. Nanoscopy, vol. 2, no. 1, p. 3, 2013.