An advanced approach for predicting selective sweep in the genomic regions using machine learning techniques

Selective sweep is an important phenomenon in the aspect of natural selection. It plays significant role in adaptability as well as survival of species, including crop cultivars. Various existing approaches for selective sweep analysis are mostly built on traditional rule base approach that lack the advanced approaches such as machine learning and deep learning and often result in poor prediction accuracy. In this study a new method or model for the prediction of selective sweep has been presented. This method has been initiated with simulation, preceded through feature extraction and selection and finally fed to different machine learning algorithms. Here eight different machine learning based methods have been implemented—(1) Support Vector Machine (SVM), (2) Regression Tree, (3) Random Forest, (4) Naive Bayes, (5) Multiple logistic regression, (6) K-Nearest Neighbor (KNN), (7) Gradient boosting and (8) Artificial Neural Network (ANN) and results of their comparative evaluations are presented. It has been observed that random forest model outperformed to its counterparts in terms of evaluation matrices with an area under the ROC (Receiver Operating Characteristic) curve (AUC) score of 0.8448 as well as 1st rank in TOPSIS (The Technique for Order of Preference by Similarity to Ideal Solution) analysis. Further, a robust model for selective sweep prediction based upon random forest has been developed. Model developed in the current study has outperformed to other existing approaches for prediction and analysis of selective sweep. This new approach for selective sweep analysis is excellent in its accuracy as well as reliability.