New Search

Dataset Listing

Show Expression Levels

About the Website

SPELL Version 2.0.2

Browser Compatibility

Some features of the SPELL website use Javascript and cookies. Please enable cookies and javascript to use the SPELL search engine.

The SPELL website has been extensively tested on a variety of web browsers running various operating systems. The following browsers are known to be compatible:

Windows: Firefox(v2+, preferred), Internet Explorer(v6+)
Apple Mac: Safari(v2+), Firefox (v2+), Other Gecko based browsers (e.g OmniWeb, Camino, etc.)
Linux: Firefox(v2+)


Algorithm Details

The main algorithm of the SPELL search engine is described in the paper: Hibbs MA, et al. Exploring the functional landscape of gene expression: Directed search of large microarray compendia. Bioinformatics, 2007.

Briefly, given a set of query genes provided the engine, a relevance weight is determined for each dataset based on how well correlated the query genes are in each dataset. Datasets where the query genes are largely co-expressed recive a high weight, while datasets where the query genes do not agree are given a low weight. Based on these per-dataset weights, weighted correlations are calculated for every other gene in the genome to the query set. In this way genes that agree with the query set in datasets where the query is consistent will achieve the best results, while genes that agree with some of the query set in datasets where the query is not co-expressed will recieve a low result. Datasets and genes are sorted by their correlation scores and returned for each query.

There are two edge cases not discussed in the main text that you may encounter:
First, if no datasets are found to contain a significant signal for a given query then we are unable to assign per-dataset weights for the search. In this case a warning message is displayed, and all datasets are equally weighted for that query.
Second, if no genes are related to the query set at a reliable confidence level, then a warning message is displayed and the confidence level is weakened until results can be obtained.
Both of these cases typically only occur when the query genes are either largerly unrelated, or highly unique. Neither of these cases occurs very often.


Implementation Details

The SPELL search engine interface was built using the Ruby on Rails framework, with a Java backend to perform the searches. AJAX elements were created using the Prototype, JQuery, Thickbox, and JTip libraries.