The purpose of the MEFIT web site is to display a set of example predictions generated by the MEFIT system using a specific collection of microarray data and functional annotations. To this end, the MEFIT site allows a user to query a biological function, after which a navigation page is shown summarizing MEFIT's predictions over our test set and unknowns (not the whole genome - yet!) within that function.
When visiting MEFIT, the first page you see is the query page. This page allows the user to select a biological function for which to view MEFIT's predictions. Once a function is selected from the drop down menu, the function navigation page will load automatically.
Alternatively, a user can filter the function list to include only those functions in which MEFIT predicts at least one functional relationship (above four standard deviations) for some gene. To do this, enter a gene name or ORF identifier in the text box and press the Query button (or the Enter key). The function list will be highlighted to indicate that the functions have been filtered for the given gene.
Once a function has been selected, MEFIT displays the function navigation page (see an example here). The purpose of this page is to present a higher confidence subset of MEFIT's predictions in a familiar format - specifically, by using MEFIT's pairwise probabilities of functional relationship as a similarity metric, we can generate a hierarchical clustering of genes within each function. On this page, you can search for individual genes or browse the microarray data provided as MEFIT's input much like you would using a program like Java TreeView or TIGR MeV. Just remember that the clustering results are based on MEFIT's test set predictions, not traditional Euclidean distances or Pearson correlations!
In addition to MEFIT's usual headers, the function navigation page contains four main parts:
-
In the upper left corner, the navigation compass (the much
larger version of this:
.
- A vertical column containing gene names and a dendrogram.
- A horizontal row containing microarray conditions.
- A collection of heat maps corresponding to microarray results.
The navigation compass is the key to interacting with the function navigation page. Clicking on its left and right arrows will browse through the microarray experiments, scrolling the column labels (conditions) left and right. Clicking on its up and down arrows browses through the gene list, scrolling the gene names and dendrogram up and down. The microarray heat maps scroll in tandem with their row and column labels. A gene query box to search for (and, if found, select) a specific gene also appears near the navigation compass.
The gene names and dendrogram are generated by hierarchically clustering MEFIT's predictions within the specified function, including only gene pairs with probabilities four or more standard deviations above average. As in most similar tree-browsing programs, clicking on a gene name will open SGD's information page for that gene. If a specific gene has been selected (either through the initial query page or using the navigation compass), it is highlighted in red.
Column labels are drawn from the authors' annotations for the 40 microarray data sets we used to test the MEFIT system. Hovering over a label will pop up the author and publication year of the appropriate microarray condition, and clicking on a label will display a summary page (including PubMed links) for the data set containing that condition.
Finally, the heat maps encode the row normalized (per data set) microarray data (log ratios for dual channel, log values for single channel) used as MEFIT's input. Yellow indicates upregulation and cyan downregulation; these are displayed at a standard -3 to 3 saturation scale.
While MEFIT itself is a general tool for microarray integration, the MEFIT web site is a little pickier about its functionality. If you experience rendering issues while browsing the site, please make sure that you are using Firefox 1.5 or greater or Safari. If this is the case and something's stilll broken, let us know!
