Using Cost-Sensitive Learning to Determine Gene Conversions

Gene conversion, a non-reciprocal transfer of genetic information from one sequence to another, is a biological process whose importance in affecting both short-term and long-term evolution cannot be overemphasized. Knowing where gene conversion has occurred gives us important insights into gene duplication and evolution in general. In this paper we present an ensemble-based learning method for predicting gene conversions using two different models of reticulate evolution. Since detecting gene conversion is a rare-class problem, we implement costsensitive learning in the form of a generated cost matrix that is used to modify various underlying classifiers. Results show that our method combines the predictive power of different models and is able to predict gene conversion more accurately than any of the two studied models. Our work provides a useful framwork for future improvement of gene conversion predictions through multiple models of gene conversion.