Alejandro1400
commented
2 weeks ago The pipeline initially groups the tracks. Obtaining:
- Mean Position: Position weighted by the quality of the localization
- Min frame, max frame
- Gaps: Frames in between which do not have localization
For the localizations it just has position, quality, intensity, snr and frame
From using ThunderSTORM we are trying to find an initial estimate for blinking statistics. Therefore we need a tracker. There are multiple Softwares specialized in SpotDetections but most of these are very basic and for the Signal to noise ratio we also need a specialized software not only for Spot detecting but also its own tracking. After initially trying using the SpotDetector and SpotIntensityAnalyzer, I turned my attention to a promising one called TrackMate:
TrackMate Paper
It has the next paper statistics:
This software is specialized for single particle tracking. In our case we want to use the automatic tracking for the molecules (It is important to remember that this has no subpixel localization).
The process for analyzing this image is displayed as follows:
This is the same region as specified in #6.
The image was left untouched and the TrackMate (In Plugins->Tracking) was selected.
Firstly, the crop settings were left untouched
The detector seleted was the Laplacian of Gaussian (LoG detector). This one allows for sub-pixel localization after searching the maxima in the filtered image and then applying a quadratic fitting scheme in the raw image
For the LoG detector an estimated object diameter in microns needs to be selected. For the selection of this parameter I used the preview feature and modified it until the molecule with the biggest area of brightness was as closely encaptured by the circle. This should better be evaluated with multiple frames.
Then, for selecting the quality threshold I can analyze the histogram and review with both this and the image what should be considered a molecule. In this case, it would be just only 4-5 molecules
After, for the initial thresholding I selected auto. This eliminates molecules for which the intensity/contrast is very low. Using auto normally doesnt take into account too many molecules, it is better to lower it until at least in the first few images the possible molecules are clearly visible. This doesn't have to be too precise since later we'll pass the images through yet another filtering step
Then, we do not apply any filtering to the steps and go to the tracking section. I applied a max linking distance and gap-closing of 0,1 micron which is the pixel size. The max frame gap of just 2 since we are not tracking molecules but events
The output displays the events as circles and the localizations as points which show the concentration of molecules. Where it becomes easier to analyze it as clusters
Finally, the idea is to click the Spots button and export it as a CSV. This includes the Spot ID, Track ID, position, intensity mean and SNR.
As said before, this doesn't show the molecules but events. A molecule can consist of one or more events in a temporal profiling. Therefore, this output will need to be further analyzed.
The molecules which don't have Track Id are those which were not assigned to an event. But still when we go through these, it is clear to see that a lot of them are part of an event but by some reason did not get assigned, the possible reasons are: 1. It had an estimated position far away from the ones in its close events. 2. It occured in a separate event. 3. It detected background as a possible molecule. For the scenario where the estimated position is far way from the ones in its close events what we would need to analyze is for the molecules part of a track if there are temporal gaps and if another event or non-track id is close enough so that we add it to the track. In the case there is a point close enough but it belongs to another track id, then it requires an additional step where we verify that there is no overlapping between the tracks. If in a same frame they both have an event then nothing should be done as it may be an scenario where simply these molecules are very close together but are separate. In the case it occurred in a separate event (it is not the possible continuation of other events) then it would need to be checked for multiple things: 1. If it has other scenarios very close both in frames and position. 2. If it has a high signal to noise ratio. 3. If it doesnt have a high SNR and none molecules close in position and frames then it should not be taken into account. 4. If it has a high SNR but it still has just one frame where it appears then it should be removed
These filtering steps are going to be done through the python pipeline, and will help then implement a similar logic but with higher precision and more specialized using ThunderSTORM