iDHS-DASTS

iDHS-DASTS identifies DNase I hypersensitive sites (DHS) from DNA sequences to detect regulatory regions involved in gene regulation.

Key Features:

• Feature Extraction with PseDNC: Employs Pseudo Dinucleotide Composition (PseDNC) to capture intrinsic DNA properties and spatial information from sequences.
• Dimensionality Reduction via LASSO: Uses the Least Absolute Shrinkage and Selection Operator (LASSO) to reduce feature dimensionality while selecting informative features.
• Stacking Learning Classifier: Implements a stacking ensemble integrating Adaboost, Random Forest, Gradient Boosting, Extra Trees, and Support Vector Machine (SVM) to combine multiple classifiers for prediction.
• Handling Imbalanced Datasets (SMOTE-Tomek): Applies SMOTE-Tomek to balance class distributions prior to training.

Scientific Applications:

• Benchmark DHS identification: Demonstrated accuracies of 92.06% (S1), 91.06% (S2), 90.72% (S3), and 90.31% on independent dataset S4.

Methodology:

Feature extraction using PseDNC, dimensionality reduction with LASSO, data balancing via SMOTE-Tomek, and classification by a stacking ensemble combining Adaboost, Random Forest, Gradient Boosting, Extra Trees, and SVM.