Among biologists, web-based genome browsers such as the UCSC Genome Browser [1] and Ensembl [2] are popular community resources for organizing and integrating diverse kinds of biological annotations and attributes that can be mapped to the genome sequence of an organism. Other graphical genome representation tools such as Apollo [3] and Sockeye [4] are popular for specialized applications in the areas of sequence annotation and comparative genomics, respectively. In addition, software such as the Generic Genome Browser, that allows individual investigators to implement their own genome browsers, has been widely used for creating browsable genome maps for diverse organisms [5]. While there are differences in representation and functionality among these genome browsers they all map genome features and their biological attributes to a common genome framework using nucleotide coordinates. The browsers and software tools listed above also share a common visualization mechanism in which different data sets are displayed as horizontal "tracks" that can be toggled on and off according to the interests and preferences of the user. The one exception to this paradigm is NCBI's Map Viewer [6] which supports the simultaneous display of maps built using different underlying coordinate spaces (genetic and genomic maps, for example) and displays maps in a vertical orientation instead of horizontally.

In geographic information systems (GIS), maps are created and displayed using 2D (or 3D) coordinate reference systems in a given coordinate space [7]. Different types of geographic features (e.g., cities, rivers, rainfall) are characterized individually and typically stored as different map layers (Figure (Figure1).1). By employing a common spatial reference system, layers are georegistered and can be overlaid on each other to visually evaluate the spatial distribution of features. Organizing data as spatially referenced layers provides flexibility to select and combine layers in various ways. Through the combination of georegistered layers geographers can evaluate what kinds of feature tend to co-localize or to explain the presence of one feature as a consequence of its spatial relationship to another feature.

In GIS, support for query and display is tightly integrated (i.e. a map is a response to a query). GIS supports spatial selection queries on individual features within a layer with the result that features meeting the query constraints are highlighted on the map. Spatial join queries are a particularly powerful GIS function that allows the user to query on spatial relations among features across map layers.

An example of a typical GIS spatial query is a query for all houses that are priced under $500,000, within School District A and less than 5 miles from a highway. This query executes a proximity query on houses and roads and a spatial containment query on houses and school district A. The result is a map highlighting only those houses satisfying the attribute constraint (price< $500,000) and the two spatial relationship constraints. An example of a spatial join query is a query for all houses with school age children and bus stops within School District A. The query returns the combined set of selected houses and bus stops falling in District A.

The Genome Spatial Information System (GenoSIS) [8] we present here adapts the GIS model to support the spatial representation of genome features. GenoSIS employs GIS functions for panning and zooming, highlighting features of interest or filtering those with certain properties; and employs standard cartographic techniques for encoding variables using graphic symbols (shape, color, etc) [9]. As in other genome browsers we define a genome map space by nucleotide coordinates along chromosomes. Unlike their treatment in other browsers, in GenoSIS, genome features are all defined as spatial objects. In this "spatial genome" representation users have the flexibility to interact with genome features as layers, as individual features within a layer, or as collections of features across layers (Figure (Figure11).

The chromosome forms the foundation layer of our implementation. Each chromosome within the layer is represented as a linear spatial object with a unique identifier and a length (in bp). The arrangement (placement and separation) of the chromosome line objects creates a 2D space. The coordinate space defined by the chromosome arrangement provides the spatial reference system and all other genome features are "georegistered" to this space.

To apply GenoSIS to analysis of the mouse genome we used genome features and attributes obtained from the following sources:

1. genes: mouse genes, their chromosome position, and the start and end coordinates (NCBI Build 34 of the mouse genome) along the genome were obtained from the Mouse Genome Informatics (MGI) database [13] public ftp site [14]

2. gene_structure: coordinate data (NCBI Build 34) for defining gene structure (i.e., intron-exon boundaries) for mouse genes was downloaded from NCBI [15]

3. GO_function: the annotation of mouse genes to high level terms in the Gene Ontology [16] were generated specifically for this study and are available online [17]

4. human_orthologs: annotations about which mouse genes have human orthologs were downloaded from the MGI ftp site [18]

5. gene_expression: a data set of developmental stage and tissue-specific expression levels for mouse genes [19] was downloaded from NCBI's Gene Expression Omnibus [20]: GDS592 [21]

6. TFBS: transcription factor binding sites used for this manuscript are for the RBP-J protein and were generated by one of the authors (CJB) using a string matching algorithm of the canonical transcription factor binding site for Build 34 of the mouse genome sequence.

• Project home page: http://www.spatial.maine.edu/~mdolan/GenoSIS.html

• Operating system(s): Free download of ArcReader for Windows. ArcReader for Linux, Solaris available for a nominal fee.