A fundamental part of this proposal is our previous work described in (Díaz and Tischer 2011). The framework developed eases the implementation of machine learning techniques for the identification of cellular automata models from protein folding trajectories. The flexibility provided by the framework facilitates the construction of applications that allow different configurations of contact maps and folding trajectory formats. This characteristic is suitable to support the exploratory research of a contact map representation for the information related to the contacts in the protein structure. So far a genetic algorithm developed with the framework has evidenced the generality of the identified CA models, opening so the possibility to ambitious proposals with practical application in an unsolved problem like contact prediction. We are aiming at the creation of a set of contact map predictors based on the previously developed framework, using alternative contact representation and new protein folding datasets.