Gene predictors ensemble for complex metagenomes

Abstract

Is presented an ensemble of predictors of genes that focuses on improving the performance of traditional predictors when applied to metagenomes obtained by sequencing 454 and are characterized by very short reads. The proposed ensemble is based on the use of data mining techniques, such as decision trees and k-means, complemented by structural information of the sequence provided by the fractal dimension. The assembly obtained can overcome the performance from the best ab initio predictor in a proportion of 15 to 20%. © Springer International Publishing Switzerland 2014.