Modeling and implementing an agent-based system for prediction of protein relative solvent accessibility

In this paper, an agent-based system for prediction of relative solvent accessibility (RSA) of proteins is proposed. Since, it is believed that the 3D-structure of most proteins is defined by their sequences, utilizing data mining methods to extract hidden knowledge and information from protein sequences, is unavoidable. Due to the inherent heterogeneity and distribution in data that used to predict RSA and high time costs of central data mining on large training data, the necessity of an agent-based architecture for predicting RSA, seems to be essential. The system is logically and functionally divided into four layers, solving the tasks of “data fusion”, “feature selection”, “model building” and “knowledge discovery and prediction”, respectively. The overall architecture has been called RSAMAS, which is implemented with multi-agent systems for prediction of relative solvent accessibility of proteins. The outcomes of the system design phase under Prometheus methodology and the complete characteristics of the agents are discussed. The prediction activity results from the interaction of a set of agents that hosted on several levels. The experiment results on Manesh dataset point to the validity of the approach. The proposed system can autonomously dig out all the valuable knowledge about which physicochemical features are highly correlated with the solvent accessibility of proteins without human supervision, which is of great importance for biologist and their future researches.