mentha: a resource for browsing integrated protein-interaction networks

Systems-level approaches require access to comprehensive 
genome-wide and proteome-wide databases. A comprehensive 
resource that archives all published protein-protein interactions 
(PPIs) is not available. In fact, primary PPI databases capture 
only a fraction of published data.

Here we report mentha (http://mentha.uniroma2.it/), a PPI resource that takes advantage of the recent establishment of the International Molecular Exchange (IMEx) 1 consortium and the development of the Proteomics Standard Initiative Common Query Interface (PSICQUIC) 2 for automatic access to molecular-interaction databases. mentha integrates protein-interaction data curated by experts in compliance with IMEx curation policies, using the PSICQUIC protocol to implement an automatic procedure that, every week, without human intervention, aligns the integrated database with data regularly annotated by the primary databases (Supplementary Methods).
The scope and motivation behind mentha are different from those of databases such as STRING, which integrate information extracted with text mining and prediction methods. mentha favors precision over comprehensiveness, and it focuses on experimentally determined direct protein interactions (Supplementary Note 1). We note that the number of interactions and proteins archived in mentha is limited by the fact that it contains data annotated exclusively in primary PPI databases, without any inference.
In designing mentha we made the following choices: (i) to focus on experimentally demonstrated physical interactions, trying to avoid confusion between physical and genetic interactions and between experimental and inferred interactions; (ii) to maintain links to original articles and primary databases; and (iii) to preserve, as much as possible, the richness of the original annotation. We restrict the integration to databases that adopt the PSI-MI controlled vocabularies 3 and the IMEx curation policies. This choice, though it excludes the use of data-rich resources that have not yet adopted the IMEx standard, such as the Human Protein Reference Database, allows for higher data consistency. As a consequence, the integration procedure in mentha can make use of specific attributes assigned according to the common curation policy, such as "interaction type" and "interaction method, " to assign a reliability score to each interaction, similarly to the Molecular Interactions scoring function 4 . The reliability score can be used to filter the PPI network of interest from Figure 1 | mentha's interactomes. The gray graph illustrates mentha's "All" interactome. The colored graphs report the interactomes of Homo sapiens and three model organisms. The insets report the number of proteins, interactions and some topological characteristics. mentha offers graph analysis tools to extract subnetworks and paths, optionally identifying enzymatic interactions.

mentha: a resource for browsing integrated protein-interaction networks
To the Editor: Systems-level approaches require access to comprehensive genome-wide and proteome-wide databases. A comprehensive resource that archives all published protein-protein interactions (PPIs) is not available. In fact, primary PPI databases capture only a fraction of published data. This dispersion of information has motivated projects such as the Agile Protein Interaction DataAnalyzer (APID), the Protein Interaction Network Analysis (PINA) platform, iRefWeb, Michigan Molecular Interactions (MiMI) and the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING), which offer wider coverage of PPI information by integrating heterogeneously curated data. The difficulty of combining annotations from heterogeneous efforts, however, consistently hampers the integration of data extracted from databases that adopt different curation policies; one consequence of laborious integration procedures is that updates are infrequent.  (Supplementary Note 2). In fact, independent experimental approaches can yield contradictory PPI information, and inconsistent data curation, or curation errors, can result in inaccurate annotation 5 . Combining the evidence from different experimental approaches can increase the confidence in any specific binary interaction. mentha archives PPI data for many species, including human; these data are updated weekly, and backups for past releases are available for download. mentha was designed as a workbench where the user can assemble and analyze collections of proteins and networks of interest ( Fig. 1 and Supplementary Note 3). mentha-the interactome browser-is accessible via a user-friendly website and via a RESTful Application Programming Interface. It also offers an interactive graphical application that can be embedded in web pages 6 .
Note: Supplementary information is available in the online version of the paper (doi:10.1038/nmeth.2561). acKnoWLedGments mentha is supported by the European Union Framework project 7 "Affinomics" grant and the "Oncodiet" grant from the Fondo per gli Investimenti della Ricerca di Base (FIRB) project.

competInG FInancIaL Interests
The authors declare no competing financial interests.

alberto calderone 1 , Luisa castagnoli 1 & Gianni cesareni 1,2
data. Previous studies from other groups have also identified discrepancies in FRAP measurements of chromatin binding that were even larger, exceeding three orders of magnitude 5 . Gebhardt et al. cited an early FRAP study of GR 4 that was based on an oversimplification of the photobleaching profile. However, the authors did not cite several later studies, one of which corrected this oversimplification 5 and another that cross-validated the newer FRAP estimate by an FCS analysis of GR binding 6 . These studies yielded residencetime estimates and bound fractions by both FRAP and FCS that are in excellent agreement with those reported by Gebhardt et al. 3 by SMT (Fig. 1). A second FCS analysis of GR 7 also yielded a bound fraction consistent with the preceding analyses (Fig. 1).
We recently reported an analogous consensus for residence times and bound fractions for another transcription factor (p53) in a three-way comparison of FRAP, FCS and SMT 2 (Fig. 1). Our p53 estimates were also consistent with an earlier FRAP analysis of p53 (ref. 8). Thus, in stark contrast to earlier measurements 4 and to Gebhardt et al.'s conclusions, agreement in live-cell binding estimates has now been obtained for two different transcription factors in five different studies using three complementary approaches. These recent data, acquired using the most current methods for performing FRAP, FCS or SMT, demonstrate that we are reaching consensus on how to extract quantitative binding estimates using not only SMT but also FRAP and FCS.

Convergence of chromatin binding estimates in live cells
To the Editor: Many transcription factors exhibit highly dynamic interactions with chromatin as measured by different fluorescence light microscopy techniques in live-cell nuclei 1 . There is mounting evidence that this transient binding quantitatively influences the process of transcription from target genes 1 . As a result, much effort has been devoted over the past decade to quantify these binding interactions, namely, to determine the fraction of transcription factors bound and their average residence time on chromatin.
Early studies were based on mathematical modeling of fluorescence recovery after photobleaching (FRAP) data, with later studies applying similar approaches to fluorescence correlation spectroscopy (FCS) data, and with the most recent analyses done by singlemolecule tracking (SMT) 1,2 . In their Nature Methods paper, Gebhardt et al. 3 developed a new strategy to improve the signal-to-background ratio in SMT of fluorescently tagged transcription factors in living mammalian nuclei, and they applied their method to measure residence times and bound fractions of the glucocorticoid receptor (GR).
The authors reported a residence time for GR that is almost an order of magnitude longer than that obtained by a FRAP measurement 4 , and they concluded that FRAP is inaccurate owing to the many complications involved in the mathematical modeling of such npg