The prediction of protein structures remains as a challenge for the scientific community. For the construction of the adequate models, several authors has been explored complex heuristic like the recurrent neural networks, multilayer support vector machines, bio-inspired algorithms or the combination of classifiers, but all these efforts are not enough. In the current manuscript the structure prediction methods based on machine learning are discussed. These methods are classified in different taxonomies and also a detailed description of each paradigm is provided, highlighting its advantages and disadvantages. Finally, the current tendencies and the last advances in the research of protein structure prediction are showed.