These results represent the primary annotations (negative weight values) and top five neighborhood weight predictions with weights above 0.025 (positive weight values) for each protein for which predictions could be made. The neighborhood weight method of annotation makes predictions based on functional categories found in proteins predicted (or demonstated in ".exp.*" files) to interact with the protein in question. These predictions are assigned a weight based on the functional quality score of the category (from automated assignment methods, manual primary annotations are given a quality score of 1.0) and the quality score of the predicted interaction. Interactions predicted based on similarity to protein pairs from the DIP or GRID are adjusted by a factor of 0.1 relative to those derived from protein pairs from the PDB. Proteins can be located based on their BioID by visiting the Bioverse (http://bioverse.compbio.washington.edu). See also eREADME.* for details about the different file types. Accuracy of the predictions can be roughly estimated based on the following table, though individual organisms vary (estimates good for top 5 predictions only). Neighborhood Weight Estimated Accuracy ---------------------------------------------------------------------- 0.5 40% 1 48% 1.5 51% 2 52% 2.5 60% 3 65% 4 70% 5 75% Accuracy can also be estimated by the ranked position of the prediction as follows: Rank Estimated Accuracy ---------------------------------------------------------------------- 1 58% 2 54% 3 49% 4 44% 5 40% Author: Jason McDermott (mcdermottj@compbio.washington.edu) Created: 1/26/2005 Last modified: 1/26/2005