Performance Analysis of Protein Contact Prediction Algorithms

Abstract

One of the most important unsolved problems in the area of Computational Biology is the prediction of protein structures. A key element in this problem is the prediction of contacts in a protein from its amino acid sequence, since it provides fundamental information for the determination of its three-dimensional structure. Due to the attention devoted to this subproblem, especially in the last decade, there are a large number of methods in the literature that obtain very good results; but there is still a considerable room for improvement. In the 13th edition of the Critical Assessment of protein Structure Prediction (CASP), a notable progress has been achieved in this area due to the use of deep learning and deep convolutional residual neural networks in state-of-the-art methods; in addition to the use of additional information from other predictions, such as solvent accessibility, conformation of the secondary structure, etc. The present work analyzes the performance of the most outstanding CASP13 methods, considering a larger test set (483 proteins) with proteins of four different classes according to SCOP. The results were evaluated using the CASP metrics. The analysis indicates that most of the selected methods have an accuracy above 90% for the test set used; SPOT-Contact being the best prediction method in general, and at least one of the best in each of the SCOP classes. The test cases and implementations made for the evaluation of results are publicly available.