96 | VOL.10 NO.2 | FEBRUARY 2013 | nature methods

correspondence

1. Csordas, A. et al. Database (Oxford) 2012, bas004 (2012).2. Vizcaíno, J.A. et al. Nucleic Acids Res. 38, D736–D742 (2010).3. Elias, J.E. & Gygi, S.P. Nat. Methods 4, 207–214 (2007).4. Gupta, N., Bandeira, N., Keich, U. & Pevzner, P.A. J. Am. Soc. Mass Spectrom.

22, 1111–1120 (2011).5. Frank, A.M. et al. Nat. Methods 8, 587–591 (2011).

incorrect identifications are evenly distributed among the target and decoy databases3. It therefore seemed conclusive that identifications with low ratios represented the random incorrect identifications of low-quality spectra.

Using our clustering algorithm, we also built a resource called PRIDE Cluster, in which we clustered all identified public spectra available in PRIDE (Supplementary Fig. 8). PRIDE Cluster con-tained 20,666,123 identified spectra from 9,040 experiments and 40 species (June 2012, Supplementary Protocol). The cluster-ing algorithm robustly organized the data in 3,152,393 clusters, a reduction by a factor of more than 5 from the original number of spectra (Fig. 1b), taking approximately 601 central processing unit days. Of the clusters, 5.7% contained spectra from multiple species that were identified as either contaminants or conserved sequences (Supplementary Fig. 9).

The PRIDE Cluster resource currently provides two main methods for accessing its data: (i) retrieve all clusters that contain a given peptide identification and (ii) retrieve all clusters with a consensus spectrum similar to a query spectrum (Fig. 1c). Researchers can now easily check whether spectra similar to their experimental ones were already iden-tified in PRIDE. As a key functionality, reliable peptide identifications based on the results from PRIDE Cluster (Supplementary Note 5) are now highlighted in the classical PRIDE web interface (http://www.ebi.ac.uk/pride/). Furthermore, PRIDE Cluster provides a simple way to correct inaccurate annotations in the PRIDE database (Supplementary Note 6). The consensus spectra of all reliable clus-ters can be downloaded as spectral libraries (http://www.ebi.ac.uk/pride/cluster/libraries/; Supplementary Table 1), which performed comparably to the corresponding ones from the National Institute of Standards and Technology (Supplementary Note 7).

PRIDE Cluster is the first step in introducing stringent quality con-trol in a highly heterogeneous MS proteomics repository and pro-vides a periodically updated (at least once a year), reliable consensus of published proteomics data.

Note: Supplementary information is available at http://www.nature.com/doifinder/10.1038/nmeth.2343.

acKnoWLedGmentsWe want to acknowledge all other members of the PRIDE team, who contributed to some aspects of this work in terms of technical implementation and discussion of the results. Additional thanks go to Matrix Science for providing a Mascot license. Special thanks go to all data submitters to the PRIDE database, whose data are the foundation of the work presented here. J.G. is supported by the Wellcome Trust (grant no. WT085949MA). J.A.V. is supported by the European Union FP7 grants LipidomicNet (202272) and ProteomeXchange (260558). J.M.F. is supported by a Biotechnology and Biological Sciences Research Council CASE studentship, also funded by Philips.

author contrIButIonsJ.G. designed and implemented the algorithm, ran the experiments, performed the analysis and developed the PRIDE Cluster application. J.M.F. contributed to the development of the algorithm and the data analysis. J.G. and J.A.V. wrote the manuscript. H.H. and J.A.V. supervised the project. All authors discussed, commented on and contributed to the final version of the manuscript.

competInG FInancIaL InterestsThe authors declare no competing financial interests.

Johannes Griss, Joseph m Foster, henning hermjakob & Juan antonio Vizcaíno

European Molecular Biology Laboratory–European Bioinformatics Institute, Wellcome Trust Genome Campus, Hinxton, UK. e-mail: jgriss@ebi.ac.uk

Accelerating 3B single-molecule super-resolution microscopy with cloud computing

To the Editor: The recently invented technique of Bayesian local-ization microscopy1 relaxes the requirement for an extremely high signal-to-background ratio for single-molecule super-resolution microscopy2–5. It enables a standard wide-field fluo-rescence microscope to generate super-resolution images from the stochastic bleaching and blinking6,7 of fluorescence dyes (3B analysis). Although the new 3B microscopy has enormous poten-tial to put the power of super-resolution imaging into the hands of average biologists, its underlying algorithm can be an impedi-ment. Calculating an area of 50 × 50 pixels takes approximately a full day on a state-of-the-art desktop computer (Fig. 1a). For larger image areas, the need for massive computing resources makes the 3B analysis inaccessible to many labs and limits its scope of application. We report here a simple solution for every-one: the Amazon Elastic Compute Cloud (Amazon EC2).

Aside from its flexible and cost-effective super-computing power, the greatest advantages of Amazon EC2 are its accessibil-ity and ease of use. Normally, porting a computational code to a new system involves the configuration of a system-dependent software environment. The convoluted configuration process—not to mention the increased complexity for cluster comput-ing—often discourages scientists without programming back-grounds. Use of Amazon EC2 significantly lowers this entry barrier. By packaging the executable program together with its running environment in a single image file known as the Amazon machine image (AMI), we can distribute 3B analysis to the public through the Amazon Marketplace. Hundreds of processes can run on instances, with each instance configured as an exact copy of the AMI. A user can launch ma ssively parallel 3B processes on the Cloud from a local computer in just a few minutes (Fig. 1b and Supplementary Video 1).

Large computing power allows data analysis without compro-mising on field of view. We tested Amazon EC2 using the photoacti-vatable (PA) mCherry–tagged podosome data from ref. 1 (Fig. 1c). The recorded image was divided into small mosaics of 12 × 12 pixels. The mosaics were analyzed in parallel and recombined to form a super-resolution image of 9.6 × 9.6 μm. Given the same computation time, an eight-thread 3.40-GHz Intel Core i7 desktop computer could analyze a region of 2.715 × 2.715 μm. The image generated from 100 virtual cores from Amazon EC2 was 12 times as large.

An entry-level user of the Amazon Cloud may obtain up to 100 instances at about $0.02–$0.06 per instance hour. Depending on the market price, reconstruction of a 25 × 25–μm image costs less than $25. We strongly recommend the use of our bash scripts to continuously monitor the reconstructed image (Supplementary Software and tutorial therein). The calculation should be

npg

© 2

013

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

nature methods | VOL.10 NO.2 | FEBRUARY 2013 | 97

correspondence

4. Rust, M.J., Bates, M. & Zhuang, X. Nat. Methods 3, 793–795 (2006).5. Huang, B., Wang, W., Bates, M. & Zhuang, X. Science 319, 810–813 (2008).6. Dertinger, T., Colyer, R., Iyer, G., Weiss, S. & Enderlein, J. Proc. Natl. Acad.

Sci. USA 106, 22287–22292 (2009).7. Burnette, D.T., Sengupta, P., Dai, Y., Lippincott-Schwartz, J. & Kachar, B.

Proc. Natl. Acad. Sci. USA 108, 21081–21086 (2011).

terminated only when the structrual features have converged, as otherwise potentially serious image artifacts may still be present. Researchers may apply for Amazon Web Service research grants for additional computing hours. We per-formed a test with 300 instances of 600 virtual cores. An image of PAmCherry-tagged tubulins in U2OS cells (from our own experiments) with the size of 150 pixels × 100 pixels × 1,500 frames could be reconstructed in only 210 min (Fig. 1d), whereas on a state-of-the-art desktop computer, this process would have taken 9 d.

Cloud computing has been transforming the information technology industry through its elastic and flexible on-demand services. Although cloud computing has contributed to the success of many companies such as Dropbox and Netflix, until now it has been little used in biological research. Through this Correspondence, we hope to highlight its huge potential for the imaging community.

Note: Supplementary information is available at http://www.nature.com/doifinder/10.1038/nmeth.2335.

acKnoWLedGmentsWe thank E. Rotsen for helping us install the 3B analysis software. This research is

supported by the Waitt Foundation.

competInG FInancIaL InterestsThe authors declare no competing financial interests.

Ying s hu1, Xiaolin nan2–4, prabuddha sengupta5, Jennifer Lippincott-schwartz5 & hu cang1

1Waitt Advanced Biophotonics Center, Salk Institute for Biological Studies, La Jolla, California, USA. 2Department of Biomedical Engineering, Oregon Health and Science University, Portland, Oregon, USA. 3Knight Cancer Institute, Oregon Health and Science University, Portland, Oregon, USA. 4Oregon Center for Spatial Systems Biomedicine, School of Medicine, Oregon Health and Science University, Portland, Oregon, USA. 5The Eunice Kennedy Shriver National Institute of Child Health and Human Development, US National Institutes of Health, Bethesda, Maryland, USA. e-mail: hucang@salk.edu

1. Cox, S. et al. Nat. Methods 9, 195–200 (2012).2. Betzig, E. et al. Science 313, 1642–1645 (2006).3. Patterson, G., Davidson, M., Manley, S. & Lippincott-Schwartz, J. Annu. Rev.

Phys. Chem. 61, 345–367 (2010).

Figure 1 | 3B analysis using Amazon EC2. (a) Computation time versus the radius of the mosaic. (b) Illustration of the Amazon EC2 cloud that contains multiple instances, each of which can be individually configured on demand. (c) Cloud computing accelerates image reconstruction with increased field of view. Left, for a measured image, the inset shows super-resolution reconstruction of an area of 2.7 × 2.7 μm using a quad-core Core i7 computer; right, reconstruction of an increased area of 9.6 × 9.6 μm using 25 EC2 instances. The improvement is 12.5-fold. (d) Top, confocal epifluorescence image of PAmCherry-tagged tubulins in a U2OS cell with a field of view of 21 × 14 μm. Bottom, reconstructed image obtained by running 300 Amazon instances for 3.5 h. The same calculation would otherwise take 9 d to complete on a single-core computer. Scale bars, 2 μm (c) and 1 μm (d).

5 10 15 20 250

5

10

15

20

25

Cal

cula

tion

time

(h)

Mosaic radius (pixels)Local

Instance

Amazon EC2

a b

c

d

9.6 μm

2.715 μm

Single B cubed Cloudy B cubed

21 μm

14 μm

ImageJ plug-in for Bayesian analysis of blinking and bleachingTo the Editor: The explosion of super-resolution microscopy research over the last 10 years has resulted in a myriad of new techniques1, all of which have distinct advantages and disad-vantages. Localization microscopy achieves an improvement in resolution by taking multiple images, each with only a few fluo-rophores emitting light2. A recently developed variation on the localization technique, Bayesian analysis of blinking and bleach-ing (3B)3, allows analysis of data with multiple overlapping fluo-rophores in every image. For those willing to invest in the longer analysis time required by 3B, this tool offers the possibility of observing cell dynamics at super-resolution, as demonstrated in the original paper.

The 3B software is currently supplied as a stand-alone program that must be compiled and installed before use, which limits its users to those reasonably familiar with programming. Meanwhile, the imaging community has greatly benefited from widely used open-source software packages, and ImageJ4 in particular has become a standard tool in laboratories that acquire microscopy images. Here we present an ImageJ plug-in for the 3B software, which includes both the 3B analysis and software for performing a reconstruction (Supplementary Software). The plug-in currently works with 32-bit ImageJ on Windows and 32- or 64-bit ImageJ on Linux. The reconstructions are themselves ImageJ images and can be manipu-lated in the standard way. This plug-in will make 3B analysis acces-sible to a far wider range of users than before and will aid the explo-ration of how structures change at the nanoscale in live cells. With a suitable sample and camera, data can be acquired using a standard wide-field system3.

The design of the plug-in allows reconstruction to be per-formed on previous runs, as the workflow in Figure 1 illustrates.

npg

© 2

013

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.