Incorporating g-gap dipeptide composition and position specific scoring matrix for identifying antioxidant proteins

Abstract

Oxidative stress can damage major cell components, including protein, DNA, lipid and cell membranes, which may make cells lose function and induce a wide variety of diseases. As an extensive kind of antioxidants in human and animals, antioxidant proteins are essential to eliminate cell damage and aging problems caused by oxidative stress. Accurate identification of antioxidant proteins is a significant step to reveal the inducement and physiological process of certain types of diseases and aging. Furthermore, newly identified antioxidant proteins may provide candidate targets for curing or alleviating diseases and slowing down the aging process. In this study, a random forest-based approach incorporating PSSM (Position Specific Scoring Matrix) and g-gap dipeptide composition is put forward to distinguish antioxidant proteins from nonantioxidant proteins. To further improve the prediction performance, the information gain combined with incremental feature selection is adopted to obtain optimal features. Compared with prior studies in testing dataset, the proposed method shows excellent predictive performance with accuracy of 0.807, MCC of 0.543, AUC of 0.939, respectively. It is indicated that this method may be an alternative perspective predictor for annotating antioxidant proteins.

Accurate Identification of Antioxidant Proteins Based on a Combination of Machine Learning Techniques and Hidden Markov Model Profiles

“…To further examine the robustness of the AOP-HMM model, we performed the same independent test on the D2 dataset built by Zhang et al [ 16 , 17 ]. The D2 dataset consists of two parts: (1) D2_train, including 100 AOPs and 100 non-AOPs; and (2) D2_test, including 74 AOPs and 392 non-AOPs.…”

“…The overall Acc reached 74.79% on the same dataset [ 15 ]. Almost simultaneously, Zhang et al put forward an RF-based method to distinguish AOPs from non-AOPs by incorporating g -gap dipeptide compositions and the position-specific scoring matrix (PSSM) [ 16 ]. Their model showed an excellent Acc of 80.7% when tested on an independent dataset [ 16 ].…”

“…Almost simultaneously, Zhang et al put forward an RF-based method to distinguish AOPs from non-AOPs by incorporating g -gap dipeptide compositions and the position-specific scoring matrix (PSSM) [ 16 ]. Their model showed an excellent Acc of 80.7% when tested on an independent dataset [ 16 ]. Subsequently, Zhang et al presented an ensemble classifier to further enhance the predictive performance of AOPs with hybrid features, including secondary structure information (SSI), PSSM, relative solvent accessibility (RSA), and composition, transition, and distribution (CTD) [ 17 ].…”

Machine learning‐based antioxidant protein identification model: Progress and evaluation

“…In addition to the above data set, there is also a data set that is widely used, collected by Zhang et al, 39 from the UniProtKB/Swiss Prot database 40 . The processing of the data set is the same as before, except that the CDHIT 41 threshold is set to 70%, resulting in lower similarity between protein sequences in the obtained data set.…”

“…This article summarizes and analyzes 11 studies 22,24,27–30,39,44–47 since 2013, of which 9 used datasets containing the data set collected by Feng et al 22 Figure 1 shows the performance comparison of the models established using the data set collected by Feng et al Five datasets used in the study included the data set collected by Zhang et al 39 Therefore, it is convenient and objective to compare and analyze the performance of various models. Figure 2 shows the performance comparison of the models established using the data set collected by Zhang et al 39 …”