Predicting translation initiation sites using a multi-agent architecture empowered with reinforcement learning

The accurate recognition of translation initiation sites (TISs) is an important stage in genome annotation. Due to the complicated nature of the genetic information and our incomplete understanding of it, TIS prediction remains a challenging undertaking. Many computational approaches have been proposed in the literature, some of which have yielded quite impressive performance. However, most of them either investigate the genomic sequences from one single perspective or apply some static central fusion mechanism on a fixed set of features. In this paper, we extend our previous work which proposed a novel multi-agent architecture for TIS prediction and explore the application of reinforcement learning into the negotiation process. Experimental results on three benchmark data sets have shown the effectiveness and robustness of incorporating reinforcement learning in the system.