Super-resolution techniques like PALM (photo-activation localization microscopy) and STORM (stochastic optical reconstruction microscopy) require accurate localization of single fluorophores detected using a CCD. Popular localization algorithms inefficiently assume each photon registered by a pixel can only come from an area in the specimen corresponding to that pixel (not from neighboring areas), before iteratively (slowly) fitting a Gaussian to pixel intensity; they fail with noisy images. We present an alternative; a probability distribution extending over many pixels is assigned to each photon, and independent distributions are joined to describe emitter location. We compare algorithms, and recommend which serves best under different conditions. At low signal-to-noise ratios, ours is 2-fold more precise than others, and 2 orders of magnitude faster; at high ratios, it closely approximates the maximum likelihood estimate.
• Software in a zip file containing:
(i) 'JDLoc_Test.m'. This script demonstrates the use of JD_2D_tuned and JD_2D_optimized.
(ii) 'genSpot.m'. A subroutine to generate an array of spot images to be used for localization. (Requires MATLAB Statistics Toolbox).
(iii) 'JD_2D_tuned.m'. Single-molecule localization using the Joint Distribution method tuned for maximum precision at low signal-to-noise ratios.
(iv) 'JD_2D_optimized.m'. Single-molecule localization using the Joint Distribution method optimized for speed and precision across a wide range of signal-to-noise ratios.
(v) 'readme.txt'. Describes the software and how to use it.
• Movie in a zip file:
Illustrates how successfully different algorithms estimate the location of a fluorophore in noisy images. Nascent RNA at transcription sites in monkey nuclei was imaged using a wide-field microscope and a CCD (90-nm pixels). Each frame in the movie shows one of 100 windows (15x15 pixels) with a signal-to-noise ratio (S:N) < 3 (value indicated in the lower-left corner), and the 2-D localizations obtained by the different methods (tuned version of JD – red circle;centre-of-mass, CM – blue dot; minimum-least-squares, MLS – orange dot; maximum-likelihood-estimation, MLE – green star). Each window is shown at low magnification in the lower-right corner to give an impression of its appearance at a typical scale. The tuned version of JD performs at least as well as the others.