One can localize better than chi-square fitting to a 2D Gaussian
Henrik Flyvbjerg, Technical University of Denmark
7 September 2010
Q: On Page 6 you write that "Once a target fluorophore is found, the signal is fitted to a 2D Gaussian distribution (or the centroid of the signal is determined)."
Cannot one do better than using a Gaussian? If not, how should the fitting be done? With ordinary least squares, weighted least squares, or maximum likelihood fitting?
A: Yes, for fluorophores with fixed orientation, a 2D Gaussian fit is not optimal at all. Better expressions for the PSF are available. For fluorescent beads a 2D Gaussian is essentially optimal. One should, however, fit with maximum likelihood, not least squares, in order to localize optimally.
See "Optimized localization-analysis for single-molecule tracking and super-resolution microscopy" by Kim I. Mortensen, L. Stirling Churchman, James A. Spudich, and Henrik Flyvbjerg Nature Methods 7, 377-381 (2010); doi:10.1038/nmeth.1447
One can localize better than chi-square fitting to a 2D Gaussian
7 September 2010
Q: On Page 6 you write that "Once a target fluorophore is found, the signal is fitted to a 2D Gaussian distribution (or the centroid of the signal is determined)."
Cannot one do better than using a Gaussian?
If not, how should the fitting be done?
With ordinary least squares, weighted least squares, or maximum likelihood fitting?
A: Yes, for fluorophores with fixed orientation, a 2D Gaussian fit is not optimal at all. Better expressions for the PSF are available. For fluorescent beads a 2D Gaussian is essentially optimal. One should, however, fit with maximum likelihood, not least squares, in order to localize optimally.
See "Optimized localization-analysis for single-molecule tracking and super-resolution microscopy" by
Kim I. Mortensen, L. Stirling Churchman, James A. Spudich, and Henrik Flyvbjerg
Nature Methods 7, 377-381 (2010); doi:10.1038/nmeth.1447
Competing interests
None declared