StackPDB: predicting DNA-binding proteins based on XGB-RFE feature optimization and stacked ensemble classifier

Abstract

DNA binding proteins (DBPs) not only play an important role in all aspects of genetic activities such as DNA replication, recombination, repair, and modification but also are used as key components of antibiotics, steroids, and anticancer drugs in the field of drug discovery. Identifying DBPs becomes one of the most challenging problems in the domain of proteomics research. Considering the high-priced and inefficient of the experimental method, constructing a detailed DBPs prediction model becomes an urgent problem for researchers. In this paper, we propose a stacked ensemble classifier based method for predicting DBPs called StackPDB. Firstly, pseudo amino acid composition (PseAAC), pseudo position-specific scoring matrix (PsePSSM), position-specific scoring matrix-transition probability composition (PSSM-TPC), evolutionary distance transformation (EDT), and residue probing transformation (RPT) are applied to extract protein sequence features. Secondly, extreme gradient boosting-recursive feature elimination (XGB-RFE) is employed to gain an excellent feature subset. Finally, the best features are applied to the stacked ensemble classifier composed of XGBoost, LightGBM, and SVM to construct StackPDB. After applying leave-one-out cross-validation (LOOCV), StackPDB obtains high ACC and MCC on PDB1075, 93.44% and 0.8687, respectively. Besides, the ACC of the independent test datasets PDB186 and PDB180 are 84.41% and 90.00%, respectively. The MCC of the independent test datasets PDB186 and PDB180 are 0.6882 and 0.7997, respectively. The results on the training dataset and the independent test dataset show that StackPDB has a great predictive ability to predict DBPs.

Prediction of Protein-Protein Interaction Sites by Multifeature Fusion and RF with mRMR and IFS

“…Performance Comparison with Existing Predictors. In previous studies, some researchers predicted interaction sites only from a single feature and method of constructing the single feature space rather than the ways used in our experiments [38,39]. To further evaluate the effectiveness of the prediction method in this work, three additional experiments are implemented to predict interaction sites by utilizing the methods of Wang et al's [8], Nguyen and Rajapakse's [27], Ofran and Rost's [6] studies and the present study.…”

BERT-m7G: A Transformer Architecture Based on BERT and Stacking Ensemble to Identify RNA N7-Methylguanosine Sites from Sequence Information

“…The stacking ensemble classifier is an ensemble method that uses the prediction results of multiple classifiers as new features for retraining. By integrating information from multiple prediction models, the stacking ensemble classifier can achieve the purpose of minimizing the generalization error and obtain better prediction performance than the single classifier [ 28 – 30 ]. In this study, we build BERT-m7G based on the stacking ensemble classifier to identify m7G sites.…”

DBP2Vec: Predicting DNA-binding proteins directly using pre-trained protein language model

“…Second, the feature extraction of these methods is complex and computationally expensive, e.g., the generation of PSSM requires PSI-BLAST [28] search in a huge protein database, the protein secondary structure and relative solvent accessibility information requires extra 2/9 prediction in advance by using specific software, e.g., SSPro, ACCPro [29]. Third, the fusion of multiple heterogeneous features may bring redundancy and noise which reduce the efficiency of the model [30]. Fourth, for the methods that use the 3D structure information, it is only applicable when the high-resolution 3D structure of protein is available, however, there only exists a small portion of samples that have 3D structure data, which will limit their further application [27].…”