Zhang, Y. et al. Protein Cell 10.1007/s13238-013-3904-1 (7 June 2013).

Microscopy techniques that rely on the localization of single fluorophores are developing rapidly, and localization algorithms are straining under the demands. One particular concern is that most algorithms assume the fluorophore dipole is freely rotating. If it is not, this assumption can result in large localization errors. Although solutions for fixed dipoles exist, there has been no solution for dipoles that are neither freely moving nor fixed. To address this shortcoming, Zhang et al. turned to a computational technique rarely used in optics: an artificial neural network (ANN). They trained an ANN on synthetic images of dipoles based on theoretical point-spread functions and then compared its performance to those of known localization methods using isotropic beads, anisotropic quantum rods and super-resolution imaging of actin and DNA. The ANN was faster and displayed tighter fits than the other methods, thus presenting a promising new tool and avenue for localization microscopy.