A computer algorithm has been described that may serve as a prototype for in silico approaches that assign function to hitherto unknown proteins. The method, described in Science (285, 751–753, 1999), is based on the observation that pairs of interacting proteins often have orthologs in other organisms in which the two proteins are fused into a single chain. Sequence homologies with different segments of a protein from another organism can therefore be used to predict an interaction between two different proteins. Applying the algorithm to the 4290 known protein from Escherichia coli , David Eisenberg and his colleagues came up with 6809 putative interacting or functional-related protein pairs in the bacteria. A similar search resulted in 45,502 similar patterns for the 5,800 yeast proteins. Although the approach yielded a high proportion (82%) of false positives, Eisenberg expects that refinements can improve the reliability of the method, particularly by eliminating homologies like nucleotide-binding cassettes. "Learning how to filter these domains out is very important," he says. Together with his colleague Todd Yeates, Eisenberg has founded Protein Pathways (Los Angeles, CA) to commercialize the method.