Przybylski, A. et al. Sci. Rep. 7, 15722 (2017). https://doi.org/10.1038/s41598-017-15313-9

Curve fitting is ubiquitous in the analysis of biological data. For example, super-resolution images reconstructed during image analysis in single-molecule localization microscopy (SMLM) can require data fitting for millions, or possibly even more, molecular localizations. Particularly for real-time experiments or for those involving very large data sets, analysis time can be a bottleneck. Pryzybylski et al. now present Gpufit, with which a general curve-fitting algorithm (the Levenberg–Marquardt algorithm) can be implemented on a graphical processing unit (GPU), speeding up data analysis substantially. The software is open source and modular and in its present version includes both least-squares and maximum-likelihood estimators. The authors report that data analysis is about 42-fold faster than equivalent analysis on a central processing unit (CPU); they reach up to 4.5 million fits per second with no compromise on accuracy. They illustrate the power of their approach on STORM data, again showing 45-fold faster analysis with Gpufit than with standard processing, with no loss of fit precision.