TargetATPsite: A template‐free method for ATP‐binding sites prediction with residue evolution image sparse representation and classifier ensemble

Abstract

Understanding the interactions between proteins and ligands is critical for protein function annotations and drug discovery. We report a new sequence-based template-free predictor (TargetATPsite) to identify the Adenosine-5'-triphosphate (ATP) binding sites with machine-learning approaches. Two steps are implemented in TargetATPsite: binding residues and pockets predictions, respectively. To predict the binding residues, a novel image sparse representation technique is proposed to encode residue evolution information treated as the input features. An ensemble classifier constructed based on support vector machines (SVM) from multiple random under-samplings is used as the prediction model, which is effective for dealing with imbalance phenomenon between the positive and negative training samples. Compared with the existing ATP-specific sequence-based predictors, TargetATPsite is featured by the second step of possessing the capability of further identifying the binding pockets from the predicted binding residues through a spatial clustering algorithm. Experimental results on three benchmark datasets demonstrate the efficacy of TargetATPsite.

ATPbind: Accurate Protein–ATP Binding Site Prediction by Combining Sequence-Profiling and Structure-Based Comparisons

“…Table 5 illustrates the performance comparison of ATPbind, ATPseq, and other existing protein-ATP binding site predictors, including three structure-based predictors (i.e., COACH, 9 3DLigandSite, 20 TM-SITEatp) and six sequence-based predictors (i.e., TargetNUCs, 26 TargetSOS, 25 TargetS, 33 TargetATPsite, 37 NsitePred, 24 S-SITEatp) on the independent test data set (PATP-TEST). Figure 3 shows the ROC curves of ATPbind, ATPseq, and the above-mentioned six sequence-based predictors.…”

“…37 From Table 1, we see that the imbalanced ratio between the number of non-ATP binding residues and the number of ATP binding residues is larger than 21. Compared to the minority class, the majority class contains lots of redundant information in the original data set, which can decrease the prediction performance and increase the training and testing time.…”

“…However, the imbalanced learning problem 36 embedded in protein–ATP binding site prediction, i.e., that the number of non-ATP binding sites is much larger than the number of ATP binding sites, is a problem that could decrease the final prediction performance and is ignored by ATPint and ATPsite. To solve the imbalanced learning problem and enhance the prediction accuracy, we recently proposed TargetATPsite 37 by integrating random undersampling (RUS) and AdaBoost 36 ensemble algorithms. Except for the above three ATP-specific predictors, many nucleotide-specific binding site predictors, e.g., NsitePred, 24 TargetS, 33 and TargetNUCs, 26 which also contain an ATP binding site prediction model, can be used to predict the ATP binding sites.…”

An ensemble classifier based prediction of G-protein-coupled receptor classes in low homology

“…Moreover, the IGA is used to optimize the parameter of PseAA which has been validated efficiently in our previous work [68]. In addition, as introduced by the recent studies [31,64,67], the ensemble classifier could achieve a higher prediction accuracy than basic algorithm (e.g. SVM [1,17,24,35], PNN [63], FKNN [20]).…”

Improving N6-methyladenosine site prediction with heuristic selection of nucleotide physical–chemical properties

“…Support vector machine (SVM), which was proposed by Cortes and Vapnik [51], has been widely used in the realm of bioinformatics [29,30,47,49,[52][53][54][55]. The basic idea of SVM is to transform the input vector into a high-dimension Hilbert space by kernel functions and then seek a separating hyper plane between classes with the maximal margin in this space.…”