Mean Reciprocal Rank

Abstract

Advances in high throughput sequencing and " omics " technologies and the resulting exponential growth in the amount of macromolecular sequence, structure, gene expression measurements, have unleashed a transformation of biology from a data-poor science into an increasingly data-rich science. Despite these advances, biology today, much like physics was before Newton and Leibnitz, has remained a largely descriptive science. Machine learning [6] currently offers some of the most cost-effective tools for building predictive models from biological data, e.g., for annotating new genomic sequences, for predicting macromolecular function, for identifying functionally important sites in proteins, for identifying genetic markers of diseases, and for discovering the networks of genetic interactions that orchestrate important biological processes [3]. Advances in machine learning e.g., improved methods for learning from highly unbalanced datasets, for learning complex structures of class labels (e.g., labels linked by directed acyclic graphs as opposed to one of several mutually exclusive labels) from richly structured data such as macromolecular sequences, 3-dimensional molecular structures, and reliable methods for assessing the performance of the resulting models, are critical to the transformation of biology from a descriptive science into a predictive science.