Measuring the Average Accuracy Performance of Homology Search

Homology searches of genomic databases are the essential step for new gene discovery. While novel search techniques are being developed to address this need, it is equally important to provide a unified experimental methodology for evaluating and comparing the different homology search methods. Current evaluation measures such as the Receiver Operating Characteristic(ROC) [5], the Coverage Versus Errors per query (CVE) plot [3], and the average precision (AP) measure [4], provide biologists with sensitivity and selectivity tradeoffs for specific queries but fail to provide an average performance analysis for typical queries. We propose a rapid and unbiased evaluation metric, recall-EPQ, to compare the average accuracy performance of different search methods.