T-3D-3D

[weidemann]

Measuring FCS with fluorescent proteins produces complex decays. For simplicity, Often I choose the T-3D-3D model with a single triplet to account for all blinking with a single time (Tau_trp) that is supposed to be still shorter than the protein diffusion (Tau_1) and couple of % for a longer diffusion term (Tau_2> ms) that covers longtail correlations of different origin (small amounts of aggregates, PSF distortions, cellular movements etc) . In aequous buffers, blinking times and the globular protein diffusion can be quite close.

I noticed many times that applying the confocal T-3D-3D model shows coupling of the triplet time (Tau_trp) and the shorter diffusion time (Tau_1). During fitting, both values run in parallel. I could not get rid of the stickyness of these two parameters. In contrast, applying the simpler T-3D model, both triplet and diffuion can be properly separated, although only a couple of µs apart.

Is there a technical reason for coupling of these two parameteres during fitting? I attach an example:
example-T-3D-3D.zip

[paulmueller]

Hey, great to hear from you.

There is a constraint in the model that forces the triplet time to be smaller than Tau_1 and Tau_1 to be smaller than Tau_2. That's it. If it gets sticky, then this is probably over-fitting (too many parameters for the given data).

Did you try a different fitting algorithm? The default is "Levenberg-Marquardt (leastsq)". I haven't looked at the example file yet.

[weidemann]

Hi Paul, thanks for immediate reply. Which algorithm would you recommend?
Hope you are fine, Thomas

[paulmueller]

I cannot really recommend any one of them. But I would try the ones mentioned in the docs:

With luck, one of the algorithms overcomes the stickiness.