This paper, concerned with the protein structure prediction problem, aims at automatically selecting the Constraint Satisfaction algorithm best suited to the problem instance at hand. The contribution is twofold. Firstly, the selection criterion is the quality (minimal cost) in expectation of the solution found after a fixed amount of time, as opposed to the expected runtime. Secondly, the presented approach, based on supervised Machine Learning algorithms, considers the original description of the protein structure problem, as opposed to the features related to the SAT or CSP encoding of the problem.