WholePathwayScope or WPS is a software tool that displays HTP data in user-defined or stored gene groups or pathways. The program incorporates a suite of pre-defined biological pathways and allows for the construction of additional user-defined pathways or collections of genes. It also allows generation of biological association networks composed of gene-pathway/term relationships, which can be further manipulated and converted to subnetworks, gene-gene, or pathway/term-pathway/term networks. Results from multiple HTP experiments can be visualized simultaneously, both as summary data from multiple pathways (WSCP) and as detailed data for individual pathways (PSCP). Results can be displayed numerically and can be color-coded according to user-defined criteria to facilitate visual analysis. The program also offers statistical evaluation of global functional category (GO term, pathway etc.) enrichment in a user's gene list, or of user-defined pattern enrichment of choice genes that have been color-coded with HTP data directly.

The program is written in Microsoft Visual Basic 6 and runs in the Microsoft Windows environment. It utilizes Microsoft Access Databases including the internal databases for gene annotations, pathways, gene ontology and disease association information as well as designated criteria (CRI) files for HTP data. Pathways and association networks are created and presented in windows or graphical user interfaces (GUIs), and stored and accessed either in individual files or dynamically within Microsoft Access Databases. Users control the program through a user interface involving GUIs provided from a series of panels, menus and windows. There is also an extensive context-sensitive help system.

WPS includes an internal database for integrating gene annotation information from both mouse and human genomes. Annotation covered includes GenBank accession numbers (GenBank IDs), Unigene IDs, Locuslink IDs (Now Entrez Gene), Gene Symbols, Aliases 31, SwissProt IDs (Protein IDs) 32, and disease information from both Genetic Association Database 33 and a partial MedGene Database 34. It also carries pathway/term information including KEGG 192235, Biocarta 36, CGAP 37 and Gene Ontology information 383940 for the purpose of gene-term association network generation and Fisher's exact test.

The overall program layout is described in Figure 1. Three types of files with different formats have been developed to display the HTP data. These three file types lay out the pathways being visualized, interconnections amongst those pathways, and data filtering and color criteria for the HTP data respectively. The first file type is the PathwayScope File (PSCP), which is graphical presentation of many metabolic pathways and gene groupings. PSCP files contain the identifiers (gene tags) for each gene in a group of related genes or in a pathway, which is colorable (see PSCP file examples in Figure 4 and 5). PSCP files can be customized and created by users or created dynamically from the internal database. For Biocarta pathways, the pathway graphs can be also visualized in a separate internet browser and data of selected genes in CRI files (see below) can be highlighted in the graphs as well (see Additional file 1 for screenshot).

The second file type is the WholeScope File (WSCP file), which is composed of a series of pathway tags. Each pathway tag in the WSCP file is linked to a PSCP file saved in a user's desktop or represents a term (either a pathway or GO term in the internal database), which is also colorable like gene tags. The PCSP file can be accessed if it is linked by a pathway tag or a new PSCP file(s) can be dynamically created if the pathway tag represents a database associated term from the WSCP file by clicking on the pathway tag. Global changes in expression levels of genes in each pathway or term can be indicated by setting criteria in CRI files (see below) to color code the pathway tags (see WSCP file example in Figure 3).

The third file type is the criteria (CRI) file, which is used to enter the user-defined color criteria and HTP data (see Additional files 35–50 for examples of raw data files used in this manuscript). Each CRI file is a Microsoft Access file that contains a HTP (e.g. microarray) dataset, the mapped gene identifiers (BaseGenBankID) for each microarray element, and the user-defined color criteria for the PSCP and WSCP files (see Additional files 19–34 for examples of CRI files used in this manuscript). HTP datasets are converted to CRI files in the program from Excel files (Microsoft, Inc) containing the HTP data through a Data Conversion Window (Fig. 2). WPS can be used for any high-throughput data as long as it is formatted as spreadsheets in Excel files and contains one of the three types of standard gene or protein identifiers including GenBank Accessions, Unigene IDs, or SwissProt IDs (see Additional file 2 or our program demo web page 41 for the file format and procedure for conversion into CRI files). For genes without available standard IDs, they can also be included into CRI files with user-assigned identifiers and used for analysis such as Pattern Extraction (see below). The HTP data in an Excel file can be an individual dataset, or combined multiple datasets in a single file (i.e. in Stanford format 42). In the latter case, the user can select an appropriate data column to build color criteria specifically for one dataset in the whole file as individual CRI file, or multiple data columns to build color criteria even for multiple datasets within a single CRI file (see the Additional file 3 for more details on the three file types in WPS).

Once a series of PSCP, WSCP and CRI files have been loaded into the program, the user has the option of proceeding along several analysis courses. Some of the features available for this analysis are described briefly below. The result section of this manuscript illustrates the program using real data examples to describe some scenarios to apply the program for data analysis. In addition, a set of tutorial movies and illustration image files may be obtained from our program demo web page 41 for many major features and general usage of the program (see additional files 15, 16, 17, 18 for some examples of demo movies).

In CRI files, users can define criteria to color code specific categories of behaviors of genes in datasets (e.g. red color for no less than 2 fold change of genes, or green color for genes flagged as down-regulated genes etc.). This kind of criteria definition can be used to extract a gene list of genes matching a specific pattern of such criteria across one or more CRI dataset files.

Pattern extraction can work in two ways: global pattern extraction across datasets, or local or pathway PSCP file scoped pattern extraction. The extracted gene list can be immediately copied and pasted to other utility windows for further analysis (see Fig. 6, 7, 8 and Additional file 4 for illustrative examples).

Using an input or filtered gene list, such as a list of genes derived from clustering analysis from other programs or pattern extraction in WPS, the associated pathways or GO terms can be identified from the internal database or user-defined PSCP files. These results are listed into the result table in a gene-term pairwise format (Fig. 8). Then, the pairwise relationships between genes and their associated pathways or GO terms in the results table can be used to generate a gene-term association network (GTAN) within a PSCP file and is illustrated in a graphical view (Fig. 9). The generation of such a network is based on the Scalable Vector Graphics (SVG) technology 43, a standard for describing the layout of two-dimensional graphics in XML. The gene-term association network can be manipulated and filtered for the purpose of different analysis needs in many ways (see a concrete example illustrated in Fig. 9, 11, 12, 14 and see Additional file 5 for more detailed description of this feature). In addition, genes associated with disease terms, which were derived from Genetic Association Database and MedGene Database and included in the internal database described above, can be highlighted and selected from the network for further analysis and network manipulation (see Additional file 6 for screenshot).

The Fisher's exact test is performed based on 2 × 2 contingency tables (whether a gene is in the given list or not vs whether this gene is associated with a pathway/term or not; see Additional file 7 for illustration of an example of 2 × 2 contingency table). Similar to EASE 44, Fisher's exact test p-values are computed for each term in a chosen system, which are then ranked from smaller to larger values, to estimate the statistical significance and enrichment of global functional categories (GO terms, pathways etc.) within a given system for a list of genes of a user's interests or that match a pattern. The biological themes of the gene list can be rapidly retrieved from GO system and Biocarta and KEGG pathway collections as top ranked terms or pathways based on the Fisher's exact test p-values (see Figure 7 and Additional file 8 for a concrete example).

In contrast to the global functional statistical estimation of a gene list, within a PSCP file being analyzed and colored by CRI file(s), the statistical significance and enrichment of genes with certain user-defined criteria, can be also estimated by Fisher's exact test for the corresponding CRI file(s) (see Additional file 9 for a screenshot illustration of this feature).

Within the information search window, one can type in a keyword (e.g. Gene Name, GenBankID etc.) to search for its relevant information, including annotation information, as well as associated disease information from the internal database. Two dataset manipulation utilities are available to conveniently manipulate the size of data files so that one or more subsets of a dataset, or sum of multiple datasets can be used for further analysis: 1. Sorting a dataset into pathway/term scoped "sub-datasets" based on PSCP files, or pathways/terms in the internal database (see Additional file 10 for screenshot); 2. Merge data files.

WholePathwayScope displays HTP data within a biological context. Figure 1 provides an overview of the basic work flow of the WPS program for data and file processing. Microarray data or other HTP data are entered into criteria (CRI) files in which the parameters for analysis are set. The data are then loaded into the PSCP/WSCP files either from the provided file collections or from the internal database, and those genes and pathways that meet user-defined criteria are flagged. In addition, gene lists can be extracted from CRI files for further Fisher's exact test analysis or creation of GTAN. Figure 2 provides an example of the data conversion window, in which gene and pathway criteria are entered. For each gene, GenBank accession numbers that correspond to the gene are collated and given a single BaseGenbank ID number.

By way of example, the WPS program was used to compare gene expression profiles between wild-type mice and two strains of genetically-modified mice that either express high levels of ABCG5 and ABCG8 (G5G8 ^Tg mice) or no ABCG5 or ABCG8 (G5G8 ^-/- mice) 4546 also see Additional file 11 for a description of material and data preparation). ABCG5 and ABCG8 encode ABC half transporters that heterodimerize to limit the intestinal absorption of dietary sterols and to promote the secretion of sterols from the liver into bile 47.

To compare gene expression patterns in the G5G8 ^Tg and G5G8 ^-/- mice and assess the reproducibility of the microarray results, microarray datasets from five expression array experiments (two from G5G8 ^Tg mice and three from G5G8 ^-/- mice) were analyzed simultaneously in a WSCP file, which includes a subset of the biochemical pathways and gene families involved in lipid metabolism (Fig. 3). If two or more genes in a pathway were significantly up-regulated or down-regulated (designated "UpHighCertainty" or "DownHighCertainty" in Table 1, respectively) for a dataset, the corresponding divided box of the pathway tag was colored red or green, respectively. Comparison of the data from the two experiments from the G5G8 ^Tg mice demonstrated consistent results for some pathways (e.g. cholesterol synthesis and bile acid synthesis) but not for other pathways (e.g. glycogen synthesis).

When the WSCP file window is displayed in the program, it is interactive. Each pathway tag in the WSCP file links to a PSCP file or a term in the internal database. For example, clicking on one of its divided boxes of the pathway tag "Cholesterol Synthesis" (Fig. 3) will open a new window with the details of the pathway, including the expression levels of the individual genes (Fig. 4). Many genes in the cholesterol biosynthetic pathway were colored red indicative of up-regulation in both datasets from the G5G8 ^Tg mice (Fig. 4). These genes were expressed at significantly lower levels (colored green) in samples from the G5G8 ^-/-, confirming that differences in expression levels of ABCG5 and ABCG8 have a significant impact on cholesterol biosynthesis in the liver. When cholesterol synthesis was measured in these two genetically-modified strains of mice, it was found to be increased in the transgenic animals and decreased in the knockouts, verifying the biochemical changes seen in these experiments 4546.

To assess the ontogeny of expression of genes in the cholesterol biosynthetic pathway, we analyzed microarray datasets from livers of wild-type mice or embryos sacrificed at different time points during development (Fig. 5). In utero, most cholesterol for the developing mouse is derived from endogenous synthesis, as reflected by the increased expression of many genes in the biosynthetic pathway at day -5 and day -3. Most of these mRNAs decreased within 24 h of birth, which correlates with the initiation of nursing. Many hepatic mRNAs in the group were increased between postnatal days 18 and 21, when the pups transition from consuming a high cholesterol diet (milk) to a low cholesterol chow diet (0.02% cholesterol by weight). During this time period, endogenous cholesterol synthesis compensates for the reduced dietary intake of cholesterol associated with weaning.

One strategy to analyze microarray or other HTP data is to look for genes with certain user-defined expression patterns across one or more datasets with some biological implications and themes. The expression level-switch phenomenon at birth and weaning across the time course experiment within the cholesterol synthesis pathway described earlier (Fig. 5), prompts us to further investigate the underlying mechanisms and relevant genes and biological processes to cholesterol metabolism. We used WPS to perform pattern extraction with a defined color criteria pattern that reflects the expression level switch around the birth and weaning time points from the time course experiments (Fig. 6). The pattern-matching panel and color template affords the user great flexibility in determining which dataset(s) to include and what color(s) to accept (Fig. 6). The resulting gene list containing 24 unique genes was copied and pasted to Fisher's exact test window to evaluate the statistical enrichment of pathways or GO terms within this list (Fig. 7; also see Additional file 8 for complete result list). As expected, the Fisher's exact test clearly indicated enrichment of sterol/lipid metabolism and synthesis GO terms in the resulting gene list (Fig. 7).

To further study the underlying relationships between genes and involved/enriched pathways or GO terms, we used WPS to search involved pathways from Biocarta, KEGG and GO/Biological Processes terms for the extracted gene list from Fig. 6, and then dynamically generated a gene-term association network (GTAN) (Fig. 8, 9). Thereby, we visualized gene-term association as well as gene-gene and term-term relationship in a graphical manner. Within a typical GTAN, any gene with an associated term has a line linked to it representative of the association relation. In this interactive window, the line linkage between the gene and its associated term will still be maintained even if the gene or the term is moved to a different location. As shown in Fig. 9, among total 24 genes from the above extracted gene list, only 14 genes have annotated association pathway/GO terms from Biocarta, KEGG and GO/Biological Processes, totalling 90 terms and 206 gene-term associations, which are included in the network (see Additional file 12 for complete pair-wise gene-term relations in this GTAN). All the genes that are highlighted in red (Fig. 9) are presented as a network along with their associated pathways/terms. At this level of a GTAN, the legends of genes and associated terms are invisible (Fig. 9). One can explore the network by moving the mouse over the top of genes or terms, the legend of which can then be displayed visibly near the mouse. Alternatively, one can "zoom in" on a selected area of the network to take a closer look at the subnetwork. There is also a specialized window available for users to manipulate, filter, and explore the network (Fig. 10).

To investigate the biological themes from the resulting GTAN, we used a variety of manipulation methods available in a utility window to filter and simplify the network within WPS (Fig. 10). In combination with the Fisher's exact test result (Fig. 7), we filtered out a subnetwork of genes with associated GO terms that have Fisher's exact test p-values no larger than 0.05 (Fig. 10, 11). Interestingly, although the majority of the enriched terms are engaged in sterol/lipid synthesis and metabolism, endocytosis and vesicle-mediated transport GO terms, which are enriched in the extracted gene list, link Ldlr and Rab4a together in the subnetwork (Fig. 11). We also retrieved terms with minimal associations with genes, which tend to be unique terms/pathways describing their involved genes (see Additional file 13). Notably, Rab4a is functionally involved in signaling and cell communication (see Additional file 13). Thus, it appears that Rab4a could be the critical signaling component that triggers and mediates the sterol/lipid synthesis machinery in the body through Ldlr , probably by the mechanism of endocytosis (Fig. 11, and see Additional file 13).

We then filtered out genes with minimal association of terms and looked for shared terms within the network (Fig. 12). As evident in Fig. 12, sterol/lipid synthesis and metabolism are major shared terms in the network, which is consistent with the filtered network from the Fisher's exact test result (Fig. 11). The fact that Hmgcs1 , Hes6 and srebf1 (SREBP) heavily shared transcription-related terms (Fig. 12), is consistent with the facts that Hmgcs1 and srebf1 have been previously shown to be involved in sterol metabolism regulation and Hes6 has been implicated in transcriptional regulation 48, suggesting that they may collaborate in metabolism-related transcriptional regulation. Egln3 only shared high level or more generic GO terms (e.g. cellular biological process) with other genes (Fig. 12), confirming its unique functional involvement in cell death among the gene list (see Additional file 13). Furthermore, Rab4a and Ldlr shared many pathway/terms, suggesting these two may be functionally coupled.

Interestingly, when we used disease-association highlight feature in WPS (see Additional file 6 for screenshot), we found that Rab4a , Ldlr , and srebf1 are all more or less associated with obesity annotated in Genetic Association Database 33 and MedGene databases 34 (data not shown).

To study how genes are related to each other directly within the network, we merged the gene-term association network in Figure 12 into a gene-gene network by their shared terms, after eliminating some generic terms for clarity purposes. The gene-gene network formed a domain-like architecture (see Additional file 14, Fig 13). The genes that are heavily involved in sterol/lipid synthesis and metabolism formed a clustered domain with massive links among them. Distinct genes with different functional trends from others tended to be separated out of the "crowded" region such as Egln3 , Hes6 , and Rab4a .

In order to learn more relationships among these "distinct genes", we highlighted them in red with WPS (see Additional file 14, Fig. 13) and then retrieved the immediate neighbors for them from the original GTAN network to create a subnetwork (Fig. 14). Since Ldlr is the only one from the "Sterol metabolism clustered domain" that connects to Egln3 , we also included it for further analysis. Notably, within the newly created subnetwork, all the selected genes are linked, although distinct domains or clusters for each gene are apparently separated very well (Fig. 14). All these domains may represent the major functional aspects underlying the "expression level-switch" phenomenon and are networked together as a dynamic biological theme. Interestingly, these genes were originally selected due to a similar expression pattern across the "cholesterol metabolism" switch points (Fig. 6, Fig. 13). Especially, Rab4a , Egln3 and Hes6 have been previously implicated in their functional categories including signaling and endocytosis 49, cell death 50 and transcription regulation 48, respectively. Whether and/or how they play roles in cholesterol metabolism related signaling, cell death and transcription regulation remains to be investigated.

The new software program, WPS, is described that facilitates and enhances the analysis of HTP data. Unique features of WPS include the ability to simultaneously display HTP data from multiple experiments within the context of known biological pathways, visualizing and analyzing gene-pathway/term and gene-gene relationships and biological implications within created gene-term association networks, extracting a gene list that may reflect certain biological themes by means of a user-defined pattern template with color cues, and statistically estimating the enrichment of biological pathways or GO terms within a distinguished list or a PSCP file under analysis (see Table 2 for comparison with several free and commercial pathway analysis tools).

WPS also interfaces easily with clustering programs by accepting the gene lists from clustering analysis. This will aid in the identification of interactions among biological pathways and relating the expression profiles of genes of unknown function to those of established pathways. The program accepts data from any microarray platform (including oligonucletide arrays and cDNA arrays), and accommodates data generated by SAGE (serial analysis of gene expression) 51 as well as proteomics data.

In summary, WPS was developed to provide the following important features which were not previously available under a single application (Table 2):

First, many of the current programs display data from just a single HTP experiment. This limitation hampers direct visual comparison of results from different HTP experiments. When the number of datasets is large, as in a typical time course experiment, it becomes much harder for investigators to remember what is happening at each time point. So far, to our knowledge, our unique way in WPS of displaying multiple datasets simultaneously is absent from most, if not all, of the free and commercial pathway-based HTP data analysis tools. Even a very large number of datasets, if pre-processed and combined for same or similar categories, can still be effectively displayed and visualized in the program. In our example, shown in Figure 5, the ability of WPS to display and analyze datasets of a time course experiment in the cholesterol synthesis pathway simultaneously is very helpful for pattern recognition, especially within the genes in this pathway at the time points of birth and weaning (Most genes tended to be down-regulated after birth and up-regulated after weaning). This is more difficult to discover in such an easy and intuitive way within a pathway without using the simultaneous coloring feature of the program. Benefited from this feature in WPS, the resulting pattern recognition in turn would lead to pattern extraction of a specific color pattern in WPS, which might have biological implications (Fig. 6, discussed below).

The second feature that distinguishes WPS from current microarray programs is the ability to display multiple pathways simultaneously, either in their entirety or in summary form. A given collection of pathway files can be grouped into a single WSCP file by means of pathway tags, and changes in the behavior of genes in each pathway can be flagged according to criteria specified by the user. Thus, the pathway(s) that are significantly affected by the experimental conditions are easily identified without having to visualize each pathway individually.

Furthermore, the generation and manipulation of biological gene-term association network (GTAN) greatly expands the capacity to study the gene-term and gene-gene relationships in a genome-wide fashion and provides a new way to look at genes and their involved pathways or functional GO terms. WPS has the statistical capacity, specifically using the Fisher's exact test method, to identify over-represented biological themes (pathways/processes/GO terms) in a given list of genes. More importantly, the filtering of the GTANs based on the Fisher's exact test result would give rise to a subnetwork enriched in genes and terms/pathways with statistical significance. This would help to narrow down the "core" genes and their associated terms/pathways with biological relevance of higher priority. A solid example of network filtering with the help of Fisher's exact test result is described in Fig. 11. Endocytosis and vesicle-mediated transport, which are enriched in the extracted gene list, co-exists and connects with the major enriched function sterol/lipid synthesis and metabolism in the filtered subnetwork. The layout of the filtered subnetwork brought these functions together with visible relations within the genes of interest (Fig. 11). Some specialized tools with more sophisticated statistical methods have been previously described including EASE 4452, Fatigo 53 and GOMiner 54, which prioritize biological themes embedded in the given gene list. However, none of these tools directly visualize and study the gene-term and gene-gene relationship within the biological contexts represented by the gene list in an intuitive graphical manner, nor do they take advantage of the statistical enrichment of terms to grasp the relations of the key genes and their associated terms/pathways like in WPS. In contrast, within a GTAN created by WPS, displaying all possible associated pathways/terms for genes of interests and allowing the dynamic layout of a network based on the current biological contexts, will allow one or more context-dependent, specific functional terms to be assigned to the gene based on its current role in such context. On the other hand, its shared terms with other genes would implicate the biological connection of this gene with its "neighbor" genes. We believe such visual cues derived from GTANs not only provide an overall biological picture of the current biological context, but also shed light on the function of a gene in the network, which might only be obvious when seen with its partners.

Currently, prediction and creation of genome-wide pathways 55, as well as utilization, and exploration of biological networks (genetic, regulation, and biochemical) as a method for data analysis is becoming a major trend in systems biology and computational biology 5657. There are many new tools and algorithms being developed to move in this direction 58596061. Many use complicated algorithms such as Bayesian network, Petri nets, probabilistic graphical model or newly defined rules to simulate regulatory networks and dynamic trajectories of genes. In addition, most of the commercial tools use gene-gene relationships as their major components of the networks, such that the users may easily lose track of the role of genes. The GTAN approach in WPS is not only simple to use, but also unique and effective in that gene-term association relationships are the major components of the network, so that users can easily keep track of genes and their involved functional terms or pathways and predict gene-gene relationships through their shared terms.

A third unique feature of WPS is pattern extraction, which is different from other pattern or profile-based approaches (e.g. typical clustering and classification methods) and some statistics-based methods (e.g. SAM 18). Instead of relying solely on data values, the pattern extraction method in WPS takes advantage of user-defined color criteria in CRI files representing HTP datasets. Since users define criteria in CRI files with logical expression and not just based on the data values of genes in datasets, genes with quite different data values can be defined as the same class of data in terms of their behaviors. For example, a gene with fold change of 2 may be defined into the same data class category as another gene with fold change of 8, if the user defines "no less than 2 fold" as up-regulated genes. If the user defined this criterion with red color, then red color would represent a class category of genes, which are up-regulated, with fold change no less than 2 fold no matter whether it is 2 fold or 8 fold change, as long as the user is confident with the definition based on his own experience. In fact, one can enhance this definition by adding other quality control factors such as p-value in the definition logical expression, another advantage of CRI color criteria. Thus, this kind of definition eliminates the mathematical difference but maintains the embedded biological meanings in the data values, since biological processes are more qualitative than quantitative in most cases.

WPS facilitates comprehensive analysis and visualization of HTP data within the context of known biological pathways and gene-term association networks. The program will continue to be improved, as characterization of biological pathways and networks becomes increasingly comprehensive and challenging. The ultimate goal of WPS is to integrate all the available information and databases as well as an individual user's data with different forms and formats in the contexts of biological pathways and networks.

The current version of WPS is a windows-based program and serves as proof of concept of pathway/gene analysis of HTP data. The future version of the program will move to a three-tier architecture in a production-scale platform to allow WPS, through a middle layer, such as Java Servlets, to communicate with the server's resource, which may have excellent data storage as well as computation capacity. Its front end interface will also evolve into a platform-independent client such as a web-browser, depending on resource performance and other factors. Integration with other data sources and additional pathways are also to be added, so that the magnitude of HTP data analysis can be largely extended with the power of an expandable server.

WPS provides Fisher's exact test method for statistical evaluation in both global system and local current PSCP files either derived from internal database or user-customized under analysis. It could be improved by the addition of more sophisticated statistical utilities such as false discovery rate (FDR) estimation 62 or other statistical enhancements (e.g. bootstrap method) to analyze HTP data and determine the significance of functional enrichment in individual pathways or GO terms in a more solid way. Computational requirements limit the full integration of these statistical methods, but even without them, our software contributes significantly to improve integrative data analysis.