An Open Source Microarray Data Analysis System with GUI: Quintet
چکیده
We address Quintet, an R-based unified cDNA microarray data analysis system with GUI. Five principal categories of microarray data analysis have been coherently integrated in Quintet: data processing steps such as faulty spot filtering and normalization, data quality assessment (QA), identification of differentially expressed genes (DEGs), clustering of gene expression profiles and classification of samples. Though many microarray data analysis systems normally consider DEG identification and clustering/classification the most important problems, we emphasize that data processing and QA are equally important and should be incorporated into the regular-base data analysis practices because microarray data are very noisy. In each analysis category, customized plots and statistical summaries are also given for users convenience. Using these plots and summaries, analysis results can be easily examined for their biological plausibility and compared with other results. Since Quintet is written in R, it is highly extendable so that users can insert new algorithms and experiment them with minimal efforts. Also, the GUI makes it easy to learn and use and since R-language and its GUI engine, Tcl/Tk, are available in all operating systems, Quintet is OS-independent too. Introduction DNA microarray is the de facto standard technology for high-throughput functional genomics in the postgenomic era [1]. Since the microarray experiment is highly evolved and requires multiple handling steps each of which is a potential source of fluctuation which undermines the reliability of the data itself [2], much effort has been exerted to understand the sources of variability and minimize them to produce high-quality reproducible data [3]. ∗ These authors contributed equally to this work. ∗∗ Corresponding author. In order for this technology to be fruitful, however, reliable analysis of the data is as important as the production of high-quality data itself. Due to the high-throughput character of the microarray data, this requires maturity in numerous statistical techniques, not to mention the data processing chores. Also required is the dexterity in scrutinizing various pertinent biological information so that one can successfully reconstruct the ‘big picture'’ of biological processes fragmentally reflected in the data. Considering these, a system that can provide analytic capability as well as informatic capability is crucial for an effective, versatile analysis of microarray data. In addition, since many new approaches appear almost daily by researchers, an ideal system should be extandable enough so that new techniques can be experimented with minimal efforts. In this article, we present an R-based unified cDNA microarray data analysis system, Quintet, the first result of our on-going project to build up a customized microarray data analysis suite. As the name suggests, th five indispensable categories of data analysis have been coherently integrated in Quintet: data processings including filtering and normalization, customized set of data quality assessments (QAs), identification of differentially expressed genes (DEGs), clustering of gene expression profiles, and classification of samples using a small set of gene expression patterns. Though many microarray data analysis systems claim DEG identification and clustering/classification the most important problems, we emphasize that data processing and QA are equally important and should be incorporated into the regular-base data analysis practices because the microarray data are quite noisy [2, 4]. Under this rationale, some set of data processing and QA procedures are implemented in Quintet and constitute the core functionality module of Quintet. Quintet is written in R which is virtually the standard platform for microarray data analysis now. Since many new algorithms are also written in R, they can be inserted into Quintet without much trouble and users can extend its functionality for their own needs. The GUI makes it easy to learn and use Quintet. Furthermore, Quintet is OS-independent since R-language and its GUI engine adapted for Quintet, Tcl/Tk, are available in all operating systems. Overview of Quintet: Data Analysis Model A simplified data analysis model we have projected in Quintet is depicted in Figure 1. In this Figure, procedures that are carried out in Quintet are depicted in colored boxes. We have not implemented any image analysis functionality in Quintet and the data analysis starts from a set of text slide data files. According to our experience, the absence of image analysis module does not cause much trouble since every scanning software has a mechanism to export slide data into text format files and detailed examination of data variables, not the visual inspection of microarray images, provide thorough understanding of microarray data. Quintet retains all the variables that scanning softwares provide for each gene since previously unused variables may turn out to be important for particular purposes, especially in QA steps. For example, a popular microarray image analysis software from Axon Instruments [5], GenePix, provides 43 variables for each gene, which enables a detailed understanding of spot intensities and their characteristics. For each slide data, we first mark genes that are doubted to be erroneous from various criteria. Then, we check the quality of each slide data using various plots and statistical summaries. The error spot flagging and QA procedures should be iterated until no further quality improvements are evidenced. We consider the interoperation of data processing and quality improvement check is very important to avoid data “overprocessing'' since any data processing can introduce unwanted artifacts which cannot be amended in downstream analysis steps. We apply normalization procedures to remaining genes according to the algorithm developed by Yang et al. [6]. This is an effort to remedy systematic artifacts that may have been introduced by signal extraction procedures. Normalized data are the basis for downstream analysis steps like DEG identification and clustering/classification. Figure 1. Simplified data analysis model. Downstream analysis steps are quite straightforward. First, DEGs are identified. Since DEGs constitute basic elements for subsequent analysis steps, reliable identification of DEGs is of utmost importance. Furthermore, since different algorithms produce different DEG sets, multiple algorithms are supplied in Quintet and users can select their own DEG set among them based on the statistical characteristics revealed by auxiliary plots provided in Quintet. Using the identified DEGs, a gene expression matrix is constructed and clustering/classification is carried out. As such, the DEG identification procedure is a dimension-reduction step in this sense. In clustering/classification, we also supplied multiple algorithms and users can experiment different algorithms to survey possible variations in clustering/classification results.