Reaction kernels: predicting enzyme functions you have never seen before

Motivation: Enzyme function prediction is an important problem in post-genomic bioinformatics. There are two general methods for solving the problem: annotation transfer from a similar annotated protein, and machine learning approaches that treat the problem as classification against a fixed taxonomy, such as Gene Ontology or the EC hierarchy. These methods are suitable in cases where the function of the new protein is indeed previously characterized and included in the taxonomy. However, given a new function that is not previously described, these approaches arguably do not offer adequate support for the human expert. Results: In this paper, we explore a structured output learning approach, where enzyme function—an enzymatic reaction—is described in fine-grained fashion with so called reaction kernels which allow interpolation and extrapolation in the output (reaction) space. A structured output model is learned via Maximum Margin Regression to predict enzymatic reactions from sequence motifs. We bring forward several choices for constructing reaction kernels and experiment with them in the remote homology case where the functions in the test set have not been seen in the training phase. Our experiments demonstrate the viabilty of our approach. Availability: Software and datasets will be made available by the time of the conference. Computing Reviews (1998)