InterProScan 5: genome-scale protein function classification.

Jones P, Binns D, Chang HY, Fraser M, Li W, McAnulla C, McWilliam H, Maslen J, Mitchell A, Nuka G, Pesseat S, Quinn AF, Sangrador-Vegas A, Scheremetjew M, Yong SY, Lopez R, Hunter S. · 2014

Read InterProScan 5: genome-scale protein function classification. by Jones P, Binns D, Chang HY, Fraser M, Li W, McAnulla C, McWilliam H, Maslen J, Mitchell A, Nuka G, Pesseat S, Quinn AF, Sangrador-Vegas A, Scheremetjew M, Yong SY, Lopez R, Hunter S. (2014) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3998142.txt.

Introduction

Philip Jones, David Binns, and a large team of co-authors describe a rebuilt version of a widely used protein tool, in Bioinformatics in 2014. The tool, InterProScan, gives a first-pass read of what a protein likely does. It runs several search programs that spot family ties, working parts, and key sites, then merges their reports.

Version five is a fresh build in the Java language. The new frame runs on multi-core machines and on clusters, so it can handle the millions of sequences that modern genome work produces. Outputs were widened and improved, with new formats and a lookup service for results already computed. Links now point from proteins to the pathways they may join. The software is free to download, and its code is open source. The rebuild aims to keep that work stable as data grows.

Catalog record

CREATORJones P, Binns D, Chang HY, Fraser M, Li W, McAnulla C, McWilliam H, Maslen J, Mitchell A, Nuka G, Pesseat S, Quinn AF, Sangrador-Vegas A, Scheremetjew M, Yong SY, Lopez R, Hunter S.
YEAR2014
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC3998142

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

InterProScan 5: genome-scale protein function classification By Philip Jones, David Binns, Hsin-Yu Chang, Matthew Fraser, Weizhong Li, Craig McAnulla, Hamish McWilliam, John Maslen, Alex Mitchell, Gift Nuka, Sebastien Pesseat, Antony F Quinn Published in: Bioinformatics DOI: 10.1093/bioinformatics/btu031 ABSTRACT Abstract Motivation: Robust large-scale sequence analysis is a major challenge in modern genomic science, where biologists are frequently trying to characterize many millions of sequences. Here, we describe a new Java-based architecture for the widely used protein function prediction software package InterProScan. Developments include improvements and additions to the outputs of the software and the complete reimplementation of the software framework, resulting in a flexible and stable system that is able to use both multiprocessor machines and/or conventional clusters to achieve scalable distributed data analysis. InterProScan is freely available for download from the EMBl-EBI FTP site and the open source code is hosted at Google Code. Availability and implementation: InterProScan is distributed via FTP at ftp://ftp.ebi.ac.uk/pub/software/unix/iprscan/5/ and the source code is available from http://code.google.com/p/interproscan/. Contact: http://www.ebi.ac.uk/support or [email protected] or [email protected] 1 INTRODUCTION The InterProScan software (Quevillon et al., 2005) is extensively used both by genome sequencing projects [Suen et al., 2011; Shulaev et al., 2011; Sato et al., 2011] and the UniProt Knowledgebase (UniProtKB) (The UniProt Consortium, 2012) to obtain a ‘first-pass’ profile of protein sequences’ potential functions. It does this by combining together search applications that predict protein family membership and the presence of functional domains and sites, summarizing their outputs. Before describing the architecture used by the new version of InterProScan, it is necessary to explain how these analysis applications work in a general sense, as it has influenced the overall design of the system. Search applications have two main modalities: The simplest [TMHMM (Krogh et al., 2001), SignalP (Petersen et al., 2011), Phobius (Käll et al., 2004)] have single algorithms that are used to give a likelihood that a particular feature exists. Other applications [Pfam (Punta et al., 2012), TIGRFAMs (Haft et al., 2012), SMART (Letunic et al., 2012), PIRSF (Wu et al., 2004), Panther (Mi et al., 2012), HAMAP (Pedruzzi et al., 2012), Prosite (Sigrist et al., 2012), ProDom (Bru et al., 2005), PRINTS (Attwood et al., 2012), CATH-Gene3D (Lees et al., 2012) and SUPERFAMILY (De Lima Morais et al., 2011)] are more complex, searching a sequence against multiple models, using a variety of algorithms [e.g. HMMER (Eddy, 2009), BLAST (Altschul et al., 1990)] and post-processing the raw output of the search algorithm before a final result is produced (see Fig. 1). Frequently, the algorithms used are computationally expensive and some kind of parallelization is necessary for large numbers of sequences to be analyzed in a reasonable amount of time. Once the search results are obtained, the InterProScan in-memory database is queried to find corresponding InterPro (Hunter et al., 2012) entries (i.e. the entries associated with a particular set of HMMs, etc) and additional database annotations, such as Gene Ontology (The Gene Ontology Consortium, 2000) terms, are associated with the results accordingly. [Figure: Comparison of the processing steps used by two different member database applications, TMHMM and Pfam] It is imperative that the framework built for managing the InterProScan analyses is highly configurable, so that the heterogeneous nature of the different applications can be represented appropriately, and the software can run in a multitude of computing environments. Previous versions of InterProScan (4.x and earlier) were difficult for users to install and configure, did not scale appropriately for the analysis of large numbers of sequences and the functionality of the software was difficult to extend. This reimplementation of InterProScan addresses the previous versions’ weaknesses and adds new features to the software. 2 SOFTWARE ARCHITECTURE The design goals for InterProScan 5 are driven by a wider set of use cases than existed for InterProScan 4 and earlier versions of the software package. In common with earlier versions, InterProScan 5 has been designed to allow the efficient characterization of relatively small numbers of sequences in a single analysis, including the ability to parallelize search jobs to minimize wall-clock time (as opposed to CPU time). New to InterProScan 5 is the ability to function on a massive scale, to allow the analysis and persistence of match data for millions of sequences under the control of a single ‘Master’ process with a high level of parallelization of the analysis steps on a cluster or supercomputer. This parallelization is done at three levels: sequence sets can be chunked into smaller sets for analysis (parallelization at the sequence level), individual analyses (e.g. Pfam, Prosite, etc) can run on separate threads on the same CPU, on different CPUs or on different machines (parallelization at the application level) and some application binaries, such as HMMER3, also take advantage of parallel computing (parallelization at the binary level). The large-scale mode makes use of a single relational database to store information about sequences (either nucleotide or protein sequences), predictive models and InterPro content, and the matches predicted by InterProScan. As InterProScan 5 makes use of the Hibernate object-relational mapping tool (http://www.hibernate.org/), it should be possible to port the back-end to run on most relational database management systems. By default, InterProScan 5 uses a pure in-memory database, which requires no configuration. 2.1 Overall system architecture InterProScan 5 has a modular Java-based architecture, which builds on best-of-breed Java technologies. InterProScan is built on a rich Java data model that incorporates mappings to both a relational database schema (using Hibernate) and a new XML schema (using JAXB). Multiple layers are build on this core, each with different functionality (e.g. persistence of the data to a relational database, running of the analyses and farming out of jobs to computational resources, see Fig. 2). [Figure: Overall system architecture of InterProScan 5] 2.2 Job management Each analysis is defined as a ‘job’ in InterPro. A job may contain any number of steps that are defined with dependencies that allow merging and branching. Both jobs and steps are defined and wired together in an XML format using a mixture of common components (e.g. modules to write out FASTA files or run binaries) and analysis-specific components (e.g. modules to parse algorithm output formats or run analysis-specific post-processing). InterProScan 5 makes use of Java Message Service (JMS) to manage communication between the components of the architecture, each of which may be run on separate physical or virtual machines (see Fig. 3). This communication system allows InterProScan 5 to be run on disparate environments, including a single machine, a local area network, a multi-core supercomputer or a managed cluster. The only requirement (at present) is that the machines share the same file system. JMS was selected after a careful review of multiple technologies for building and running distributed systems. Other evaluated technologies included MPI and Hadoop (http://hadoop.apache.org). MPI, although mature, did not have a well-supported official Java binding and Hadoop required a set up that was not compatible with all InterProScan use cases. JMS has proven to be robust, scalable and reliable for inter-process communications and is a stable industry standard (JMS version 1.1 was finalized in April, 2002). InterProScan 5 makes use of the open source Apache ActiveMQ JMS implementation (http://activemq.apache.org/). [Figure: Use of JMS to manage allocation of jobs across a compute resource. This figure shows the

Read the complete text