1. Introduction

Now, most of internet protein data storages provides some search tool. Typical parameters are Identifier, Pubmed Id, Description, Keywords, Release Date, some numerical values. At the same time, one cannot search by calculation-depending parameters, such as Root Mean Square Deviation between different conformations of one structure. This value, while is not optimal measure of structural differences in some cases, still is very important in almost all fields of structural biology of proteins. For instance in[5] authors tried to compose a path through multiple conformations of Apo-Calmodulin on the basis of RMSD ordering. So, we had to develop service for easy searching structures by this parameter. Distance between conformations serves as crucial factor for modelling technique choice when it is necessary to build a trajectory of protein’s behaviour. Sometimes it is possible to use various local tools for RMSD calculation (for instance Trajectory Tool in VMD package[6]). But these tools are not able to select structures from PDB.org with RMSD required. The only online service and database that provide a kind of search by RMSD is Protein Conformation Database, PCDB[7]. Although PCDB provides some kind of RMSD selection in our opinion our tool offer more quick, clear and straight search directly in PDB database.

PDB-by-RMSD is a tool that provides a simple and easy-to-use interface for searching of protein structures in the PDB archive[8] by their RMSD. Search can be performed by several parameters but the main purpose of the tool is to provide structures selected by RMSD range specified by users.

The aim of the paper is to describe the new search tool PDB-by-RMSD. We describe the core organization of this tool and its functionality. Then we explain fields of the form for requests and show how PDB-by RMSD solves user’s requests.

2. Materials and Methods

PDB-by-RMSD has been written using the C# programming language and .NET Framework platform. Following libraries has been used: Microsoft .NET Framework 4.0, Math.NET Numerics, zlib.net and Google protobuf. Source code and binaries are available on demand.

The PDB-by-RMSD application supports work with the PDB.org archive and its copy on a local machine.

The source PDB archive located on the PDB.org server takes about 160 Gb, so one-by-one entities processing could take a lot of time (because PDB service doesn’t support RMSD value as query parameter and each entity has to be downloaded on local machine to be processed). Therefore a local database was developed. It stores all PDB files in a compressed format and contains all required information of the original database such as keywords, atom coordinates, etc. Due to the local databases structure the fully updated local database takes about 50 Gb.

The local database supports synchronization with PDB.org and other databases. This functionality is provided by plugins.

Basically, any external database engine could be used as database (for instance, organization’s SQL server). Core library contains set of public .NET interfaces (CommonLib. Querying. IDataLoader and CommonLib. Querying. Idata Saver), which has to be implemented and compiled as dynamic libraries to provide such functionality. One doesn’t need to recompile core libraries or use its source code to add new data providers.

Current implementation of local database keeps structures in binary format. Besides, all information that could be used for fast search (Id, keywords, RMSD values) is in one small index file.

The PDB archive updates on Wednesdays so you should update your local database on these days to use actual data.

PDB-by-RMSD service interacts with Protein Data Bank using its SOAP service. Plugin system was developed due to flexibility purposes. Plugins can add new RMSD minimisation methods or new data source providers. Local data base is also controlled by plugin, so it can be replaced by any relational database.

We use a simple and fast algorithm of structure alignment to avoid possible solid-state transitions and rotations. The web-service minimizes RMSD by the method based on three point superimposition and quaternions[9]. Then PDB-by- RMSD calculates various types of RMSD, such as full-atomic distances with or without hydrogen, Cα trace RMSD, main chain RMSD with or without oxygen. These values are indexed and stored in local tables for very fast solving requests and reporting.

3. Results and Discussion

The developed tool is available in two types as web-service and as standalone application.

The web-service is a part of our laboratory site and available at http:// bioinfolab. ifmo. ru/ Services/ Show Service RMSD. It provides an ability to search structures in the PDB archive by the following parameters: RMSD, Structure ID, Expirement Type, Keywords. The standalone application is also available.

By clicking on the structures ID you will be redirected to the PDB.org page containing detailed information about this structure. The RMSD value is calculated only for structures whose information was obtained by the NMR Experiment Type.

All requests and their results are cached, so you can get results for common requests fast enough. Having pressed the Calculate button, you will be redirected to the URL which uniquely defines your request. Opening this URL on any computer the same page with cached request results will be displayed. It can be useful for complicated requests that can be processed for a long time because you don’t need to keep your browser open. Remember the requests URL and repeat it later to see your results.

The table 1 contains descriptions of the search parameters, their available and default values.

Table 1. Search parameters. User can specify experimental method, the range and type of RMSD, desired number of structures and keywords

Graphic user interface supports two languages: English and Russian. This tool includes dynamic library and provides the following functionality:

• API to work with the Protein Data Bank archive

• API to calculate and minimize RMSD

• Creation of a compressed local database local database

• API to create plugins with new RMSD minimization methods

• API to create plugins working with custom databases

• Saving of calculated results in the MS Excel file

The search parameters are the same as in the web interface, except for Choose by and Converters.

The Choose by parameter defines a way of selection: aphabetical or random. The Converters list represents selected minimization algorithms.

Search results are displayed in the left pane of the main dialog. To calculate RMSD you should select one of them, define the RMSD Type parameter and click Calculate.

The Show details checkbox is used for switching between RMSD calculation results representation. If it is unchecked the highest RMSD value in all frame combinations will be displayed. If it is checked all frame combinations and RMSD values between them will be displayed. The highest RMSD value is used for searching.

Suppose you need to find some structure for visualizing protein dynamics[1]. There should be structures with RMSD values in certain range, for example from 3 to 5 angstroms. To get them you should go to site and set next values:

• Min RMSD — 3

• Max RMSD — 5

• Expirement Type — NMR (to filter out structures without RMSD)

• Count — 10 (to limit count of result structures)

Figure 1. Left form contains example of search request. In this case user is going to find 10 entities with RMSD from 3 to 5. Query results are on the right pane

Click Calculate. After short delay you’ll see web page with next structures: 2A36, 2A37, 139D, 2A3J, 143D, 2A51, 186D, 1A11, 1A13 and 1A1T (figure 1).

Consider the first one, 2A36. The maximum RMSD value calculated by VMD[6] is 3, 459, as well as value on the web-page. Our other tests show that PDB-by-RMSD provides same values of RMSD as RMSD Trajectory Tool plugin for VMD. The lists of results were composed properly according calculated RMSD.

The average time of reply on request of 10 structures with RMSD specified is 7-10 seconds. It may be different if the number of structures required changing.

4. Conclusions

It is obvious today that not only structure determines function of proteins but also their behaviour[10]. This consideration results in development of numerous techniques for protein motions modelling. Methods and modifications of molecular dynamics (MD) are one of well known and well developed. At the same time MD still has numerous of limitations of implementation – time scale, number of atoms, local energy barriers and calculation expenses. In order to override these limitations lot of coarse-graned models were developed[11-14] as well as field of so called scientific visualization and animation. Both MD approaches, coarse-graned and animation approaches are very sensitive to distances between conformations of the protein of interest. Even choice of a method to apply is related with RMSD. The quality of results obtained is under in under influence of structural differences. Another motivation of this work is absence of RMSD information in .pdb records and impossibility to run such searches easily.

In this work we introduced the new service for fast search of proteins by RMSD between conformations. By applying a local database and indexing search requests are processed very quickly. Many types of RMSD are available for user as well as keyword filtration. In many fields of structural modelling from molecular dynamics to scientific animation the distance between models is very important for method choosing and analysis of results. This new instrument for PDB search by RMSD criterion is available for scientific community.

ACKNOWLEDGMENTS

We are grateful to Anton Neyolov for reading this paper and for advices.

This work was supported by the Ministry of Education and Science of Russian Federation (federal special purpose program of Russian Federation "Research and development on priority directions of scientific-technological complex of Russia for 2007-2013" № 14.514.11.4068).