Skip to Main Content
Accurate translation initiation site (TIS) prediction is very important for genomic analysis. It is a com- mon understanding that analyzing the large amount of genomic data by pure biological methods is impracti- cal if not impossible. Therefore many approaches have been proposed which apply some machine learning tech- nique to analyze a particular aspect of the data. We believe, however, that taking one single measure on the genomic data of intrinsically complicated nature will not yield a very comprehensive analysis. In this paper, we support this argument by showing how a particular biological measure is good for TIS prediction in certain sequences, but not others. We propose a novel approach which uses multiple agents, each of which investigates some distinct biological perspective. Since it is not al- ways necessary to involve all the agents to analyze any given data set, we introduce a heuristic component to predict the most appropriate agent combination scheme for the data given. Experimental results on two bench- mark data collections demonstrate the applicability and effectiveness of our proposed approach. Keywords: translation initiation site prediction, multi-agent system, ribosome scanning model, gene ex- pression data.