uncertainty filtering for different fluorophores

106 views
Skip to first unread message

miinao...@gmail.com

unread,
Mar 26, 2018, 9:10:26 AM3/26/18
to ThunderSTORM: discussion group
Hi,

I've heard that filtering of the results based on the uncertainty value should be done to all the images and the treshold value for uncertainty depends on fluorophore. For Alexa Fluor 647 I have filtered the data followingly: uncertainty < 15.0. However, I have not managed to find any literature about this issue, i.e. which uncertainty tresholds should be used for different fluorophores. Could you help me with this? I'm currently setting up a multicolour STORM system and have to find a way to treat the data from different fluorophores.

Thanks a lot!

-Miina

Brian S

unread,
Apr 2, 2018, 2:53:55 PM4/2/18
to ThunderSTORM: discussion group
Hi Mina,

Just my two cents, but I would argue that an acceptable uncertainty range is probably best determined empirically on your system as opposed to referencing literature values.  Localization uncertainty is calculated from the photon and PSF fit parameters in your experiment (you can find the equations here), so it reflects many factors that are specific to your instrument and detector.  Literature values can give you a decent range of what is possible, but optimal filter settings may be better determined by scrutinizing the properties of your own data.

Thankfully this can be done easily in ThunderSTORM.  One of the useful functionalities built into the software is the ability plot a histogram of your data (in this case from the 'uncertainty [nm]' column), and then apply a filter based on an ROI you choose in the histogram image (using the rectangle selection tool in ImageJ) or manually entered into the filter box in ThunderStorm.  This way, your selection of upper and lower bounds on uncertainty will be determined from a distribution of your own data.  You can also see the statistics on your data within the histogram window.

Good luck,

Brian

miinao...@gmail.com

unread,
Apr 3, 2018, 12:25:44 PM4/3/18
to ThunderSTORM: discussion group
Hi Brian,

and thanks a lot for your answer! I was also thinking the option of determining the uncertainty level based on the data, so after your recommendation I will definitely try that. However, if I want to do some quantitative image analysis and compare different samples, I have to use the same level throughout the whole sample set to keep them comparable. But I guess I can just evaluate several good images of certain fluorophore and imaging buffer, and then end up using a fixed uncertainty level value obtained from these images for all the images with this same fluorophore in the same buffer, independent of the target. Do you agree?

Do you have any rule of thumb how you choose the ROI? Or any statistical justifications, e.g. 95% of the datapoints have lower values than the chosen level etc.?

Thanks again!

-Miina

Reply all
Reply to author
Forward
0 new messages