使用增強學習法結合不同基因搜尋程式以提高基因識別之準確率

Abstract

A great number of gene-finding programs have been developed for annotating newly sequenced DNA genomes. However, none of them have consistent performance over various species. Recently, a decision fusion concept that improves the prediction accuracy by combining the predictions obtained by multiple gene-finding programs has been raised. The existing combination methods are relatively ad-hoc or lack intensive experiments. In this paper, we propose a new combination method based on reinforcement learning which learns from history predictions obtained by existing gene-finding programs and derives the optimal policy for selecting the best prediction program at each nucleotide. The experimental results manifest that the proposed method can significantly improves the performance compared to the single best program.