Enhanced Prediction of plant virus-encoded RNA silencing suppressors by incorporating Reduced Set of Sequence Features using SMOTE followed by Fuzzy-Rough Feature Selection Technique

Plant viruses are observed to be the natural focuses of RNA silencing. Highly diverse silencing suppressor proteins have been developed by them due to their counter defensive strategy. Silencing suppressor proteins groups share very low sequence as well as structural similarities among them. Therefore, these proteins obstruct their annotation using sequence similarity-based search techniques. Machine learning techniques can offer an alternative and effective way to solve this problem. However, the optimal performance through machine learning based techniques is being predominantly affected by diverse factors, such as availability of irrelevant and/or redundant features, class imbalance, and selection of suitable learning algorithm. In the current study, we present a novel approach to improve the prediction performance for the RNA silencing suppressors by using fuzzy rough feature selection technique with rank as well as evolutionary search on optimally balanced dataset. From the experimental results, it is obvious that fuzzy rough feature selection technique with evolutionary search on optimally balanced dataset by Synthetic Minority Over-sampling Technique (SMOTE) produces the best results with the sensitivity of 98.90%, specificity of 95.30%, overall accuracy of 96.60%, AUC of 0.993, and MCC of 0.934 using boosted random forest algorithm. On the basis of conducted experimental results, it can be observed that the proposed technique is producing the best results till date. These results can be achieved by suitably modifying the class distribution by using SMOTE followed by choosing the relevant and non-redundant features from training sets using fuzzy rough feature subset selection with evolutionary search.