Protein contact map prediction using multi-stage hybrid intelligence inference systems

• Published in: Journal of biomedical informatics Vol. 45; no. 1; pp. 173 - 183
• Main Authors: Abu-Doleh, Anas A., Al-Jarrah, Omar M., Alkhateeb, Asem
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.02.2012
• Subjects: Algorithms; Amino Acid Sequence; Amino Acids - chemistry; Databases, Protein; Fuzzy inference system; Fuzzy Logic; Hybrid prediction model; Protein Conformation; Protein contact map prediction; Proteins - chemistry; Hybrid prediction model; Fuzzy inference system; Protein contact map prediction
• ISSN: 1532-0464; 1532-0480
• DOI: 10.1016/j.jbi.2011.10.008

[Display omitted] ► JUSTcon predictor consists of multiple parallel stages of ANFIS and KNN classifier. ► A smart filter was applied to ensure the normal behaviors of the predictions. ► JUSTcon’s average accuracy was 45.2% for the sequence separation of six amino acids. ► JUSTcon outperformed SVMcon and PROFcon in the cases of large separation distances. ► For CASP9 targets, the average accuracy was 15% for sequence separation ⩾24. Proteins are one of the most important molecules in organisms. Protein function can be inferred from its 3D structure. The gap between the number of discovered protein sequences and the number of structures determined by the experimental methods is increasing. Accurate prediction of protein contact map is an important step toward the reconstruction of the protein’s 3D structure. In spite of continuous progress in developing contact map predictors, highly accurate prediction is still unresolved problem. In this paper, we introduce a new predictor, JUSTcon, which consists of multiple parallel stages that are based on adaptive neuro-fuzzy inference System (ANFIS) and K nearest neighbors (KNNs) classifier. A smart filtering operation is performed on the final outputs to ensure normal connectivity behaviors of amino acids pairs. The window size of the filter is selected by a simple expert system. The dataset was divided into testing dataset of 50 proteins and training dataset of 450 proteins. The system produced an average accuracy of 45.2% for the sequence separation of six amino acids. In addition, JUSTcon outperformed SVMcon and PROFcon predictors in the cases of large separation distances. JUSTcon produced an average accuracy of 15% for the sequence separation of 24 amino acids after applying it on CASP9 targets.