HTSanalyzeR: an R/Bioconductor package for integrated network analysis of high-throughput screens.

MOTIVATION: High-throughput screens (HTS) by RNAi or small molecules are among the most promising tools in functional genomics. They enable researchers to observe detailed reactions to experimental perturbations on a genome-wide scale. While there is a core set of computational approaches used in many publications to analyze these data, a specialized software combining them and making them easily accessible has so far been missing.
RESULTS: Here we describe HTSanalyzeR, a flexible software to build integrated analysis pipelines for HTS data that contains over-representation analysis, gene set enrichment analysis, comparative gene set analysis and rich sub-network identification. HTSanalyzeR interfaces with commonly used pre-processing packages for HTS data and presents its results as HTML pages and network plots. AVAILABILITY: Our software is written in the R language and freely available via the Bioconductor project at http://www.bioconductor.org.

1 INTRODUCTION

In recent years several technological advances have pushed gene perturbation screens to the forefront of functional genomics. Combining high-throughput screening (HTS) techniques with rich phenotypes enables researchers to observe detailed reactions to experimental perturbations on a genome-wide scale. This makes HTS one of the most promising tools in functional genomics.

Although the phenotypes in HTS data mostly correspond to single genes, it becomes more and more important to analyze them in the context of cellular pathways and networks to understand how genes work together. Network analysis of HTS data depends on the dimensionality of the phenotypic readout (Markowetz, 2010). While specialized analysis approaches exist for high-dimensional phenotyping (e.g. Fröhlich ), analysis approaches for low-dimensional screens have so far been spread out over diverse softwares and online tools like DAVID (Huang ) or gene set enrichment analysis (GSEA; Subramanian ).

Here we provide a software to build integrated analysis pipelines for HTS data that contain gene set and network analysis approaches commonly used in many papers (as reviewed by Markowetz, 2010). is implemented by S4 classes in R (R Development Core Team, 2009) and freely available via the Bioconductor project (Gentleman ). The example pipeline provided by interfaces directly with existing HTS pre-processing packages like cellHTS2 (Boutros ) or RNAither (Rieber ). Additionally, our software will be fully integrated in a web-interface for the analysis of HTS data (Pelz ) and thus be easily accessible to non-programmers.

2 AN INTEGRATED ANALYSIS PIPELINE FOR HIGH-THROUGHPUT SCREENING DATA

takes as input HTS data that has already undergone pre-processing and quality control (e.g. by using cellHTS2). It then functionally annotates the hits by gene set enrichment and network analysis approaches (see Fig. 1 for an overview).

Fig. 1.

takes as input HTS data that has already been pre-processed, normalized and quality checked, e.g. by cellHTS2. then combines the HTS data with gene sets and networks from freely available sources and performs three types of analysis: (i) hypergeometric tests for overlap between hits and gene sets; (ii) gene set enrichment analysis (GSEA) for concordant trends of a gene set in one phenotype; (iii) differential GSEA to identify gene sets with opposite trends in two phenotypes; and (iv) identification of subnetworks enriched for hits. The results are provided to the user as figures and HTML tables linked to external databases for annotation.

Gene set analysis: implements two approaches: (i) hypergeometric tests for surprising overlap between hits and gene sets, and (ii) gene set enrichment analysis to measure if a gene set shows a concordant trend to stronger phenotypes. uses gene sets from MSigDB (Subramanian ), Gene Ontology (Ashburner ), KEGG (Kanehisa ) and others. The accompanying vignette explains how user-defined gene sets can easily be included. Results are visualized as an enrichment map (Merico ).

Network analysis: In a complementary approach strong hits are mapped to a network and enriched subnetworks are identified. Networks can come from different sources, especially protein interaction networks are often used. In we use networks defined in the BioGRID database (Stark ), but other user-defined networks can easily be included in the analysis. To identify rich subnetworks, we use the BioNet package (Beisser ), which in its heuristic version is fast and produces close-to-optimal results.

Comparing phenotypes: A goal we expect to become more and more important in the future is to compare phenotypes for the same genes in different cellular conditions. supports comparative analyses for gene sets and networks. Differentially enriched gene sets are computed by comparing GSEA enrichment scores or alternatively by a Wilcoxon test statistic. Subnetworks rich for more than one phenotype can be found with BioNet (Beisser ).

3 CORE CLASSES AND METHODS

The two core S4 classes in are ‘GSCA’ (Gene Set Collection Analysis) and ‘NWA’ (NetWork Analysis). S4 methods for both classes cover the following functions:

Preprocessing: The S4 methods ‘preprocess’ reformat the input data, e.g. by removing duplicated genes and converting annotations to Entrez identifiers. This step makes the objects of class ‘GSCA’ and ‘NWA’ ready for the following analyses.

Analyses: The S4 methods ‘analyze’ are provided for gene set and network analyses. Each method depends on several input parameters which can be defined by the user. also implements a standard analysis option using default parameters that we have found to work well in many applications.

Visualization: GSEA random walks, enrichment maps and rich subnetworks can be viewed by S4 methods ‘viewGSEA’, ‘viewEnrichMap’ and ‘viewSubNet’, respectively.

Reporting: The analyses results of class ‘GSCA’ and ‘NWA’ can be reported seperately or together to HTML files using the S4 methods ‘report’ and ‘reportAll’, respectively. The output format was inspired by cellHTS2 and contains network figures as well as tables linked to external databases.