Introduction

The BLAST program is at the base of a plethora of sequence similarity studies from genomics to proteomics. Ortholog-based function prediction [1], detection of the presence of similar domains in different proteins [2] and of common exons among alternatively spliced isoforms [3], [4], analysis of the presence of common regulatory signals in potentially functionally related proteins or genes [5], [6] are just some of most common problems that can be answered by carefully analysing the results of different database searches.

BLAST related tools are available both for downloading [7], [8] and by web applications [9], [10]. While the former is a more flexible scenario and allows users to perform searches against local customized databases, the latter usually just permits to query available databases (e.g. nr, uniprot, ESTdb, etc.), but generally provides a more structured result report. The first solution generally requires additional packages or ad-hoc solutions to manage and post-process BLAST results especially for long and complex queries. Several stand-alone tools are available (some examples are: Bioperl [11], Moulder [12], Zerg [13], Museqbox [14]). They run locally, offer both simple and complex ways of filtering results (mainly on the basis of parameters such as as e-value, similarity, etc.) and return parsed results as flat files or in a spreadsheet format. These packages generally have good performances in terms of speed and flexibility but require a certain level of computer knowledge. They have to be installed and often work in a command line fashion. Alternatively, some stand-alone tools as NOBLAST-JAMBLAST [15] and Batch Blast Extractor [16] provide a graphical interface to manage the results; they also run locally, are platform independent and do not require any computer knowledge. NOBLAST-JAMBLAST requires a MySQL server to be installed. This allows the user to sort and filter BLAST results and to perform queries on them via a graphical interface. An additional class of programs (BOV [17], PLAN [18], nuclearBLAST [19]) are available via web (or require a web server installation on Linux platform), let the users upload their own BLAST results as plain text files, and provide a user-friendly environment to filter and analyse the results. These applications generally work by storing BLAST results in a relational database, which, contrary to the case of NOBLAST-JAMBLAST [15] is not accessible to the user.

Available packages and web services are designed to simplify the use of database search tools for both programmatic and manual usage and all of them have specific features that make them appropriate for specific applications. However, so far, no tool is available for reciprocal blasting of two user defined sequences dataset and for an easy mining of results in an efficient and user friendly fashion. Yet, this is not an uncommon need in biology [20]–[25].

This led us to develop a powerful tool, named PARIGA (it comes from the Sardinian idiom: “a pair of”), that, given two user selected protein and/or nucleotide datasets, performs all-vs-all reciprocal BLAST searches [7] using each member of the first set as a query and the second set as the target database and vice versa and permits to quickly filter and analyze the results. It enables the user to interrogate and compare intra- and inter- BLAST results by performing logical operations on them. Moreover, the web implementation allows directly the parsing of BLAST results. The action of filtering is performed in real-time and it is completely reversible since it acts at level of results visualization; thus, it does not require additional computational steps and it does not provoke loss of data at each filtering phase.

Methods

In our system, the BLAST outputs are stored in a python-serialized-objects relational database by a dedicated version of a python package named BlaSTorage [26]. In short, the system works by generating and manipulating a universal data structure, a JSON (JavaScript Object Notation) object, which is a lightweight data-interchange format, for each of the BLAST outputs generated with the Blast+ implementation [27]. The JSON object does not contain the alignments (the largest part of the outputs), while a dynamic generation of javascript code is used to keep the size of these objects rather small. Furthermore, searches, ranking and filtering operations can be performed on the client rather than on the server side thereby substantially improving the speed of response. In other words, the technical implementation of the system is such that the interactions between the user and the server are kept to a minimum. Although, in principle, the allowed size of the dataset is unlimited, in the current release we encourage an upper limit on the size of the two input files (an amount of 1000 proteins of about 300 residues each corresponds to about 100 MB of transferred JSONs), but it should be noted that this is only due to the limitations in the browser capability of displaying large amount of data (client-side), rather than to the computational load (server-side).

Results and Discussion

The two datasets can be uploaded by the user who can select the desired traditional BLAST parameters (word size, expected e-value, gap-open and extension, etc.). The visualization of each of the BLAST results is very similar to the standard tabular BLAST output with the possibility of retrieving the alignment by clicking on the entry name. Summary table, statistics and a graphical summary of matches can be viewed by clicking on the appropriate icon in the table header. Two additional columns have been added to the traditional schema, named coverage and inv-cov. These indicate the fraction of the query and the subject involved in the alignment, respectively. Two novel key functionalities have been implemented. First of all, the results can be visualized and filtered in real time according to parameters such as similarity, coverage, e-value etc. (Figure 1, left panel). Second, since results are stored in a hidden database structure, the user can easily perform logical operations on them by simply selecting one of the options COMMON, CROSS and MULTIPLE (Figure 1, right panel, and schema in Figure 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Real Time Filtering and Logical Operations. (Left) Real Time Filtering.

Differently from other available tools, filtering of the returned results can be done in real time with PARIGA. By clicking the filter button, a form will appear where the user can insert the desired values (or ranges) and only filtered results will be shown. Four icons on the table header will show, from left to right, a graphical summary of the hits distribution on the query sequence, the results table, a Blast summary table and the Blast statistics. (Right) Logical Operations. The main result page will show three buttons that allow the user to perform logical operations between the two groups of results as described in the text.

https://doi.org/10.1371/journal.pone.0062224.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Pariga logical schema.

Central columns represent the original input files, while results are indicated in the columns on the side. Boxes indicate logical operations that can be performed on the results. As an example: COMMON: which sequence(s) of the dataset B is(are) shared in BLAST results of sequence A2 and A3 of the dataset A? CROSS: once sequence A1 is selected from dataset A, in which results of the dataset B does it appear? MULTIPLE: which sequence of dataset A appears more than once (i.e. matches more than one region) in the results of sequence B2 of dataset B?

https://doi.org/10.1371/journal.pone.0062224.g002

Supporting Information