UMAP-DBP

UMAP-DBP predicts DNA-binding proteins from primary protein sequences using feature extraction and machine learning to support identification of DNA-binding proteins in genomic research.

Key Features:

• Feature Extraction: Extracts features from primary protein sequences using Physicochemical Distance Transformation, Profile-based Auto-Cross Covariance (PseAAC), and General Series Correlation Pseudo Amino Acid Composition (GSC-PseAAC).
• Dimensionality Reduction: Applies Uniform Manifold Approximation and Projection (UMAP) for reducing feature dimensionality.
• Feature Importance: Utilizes feature importance scores in conjunction with UMAP to refine selected features.
• Classification Algorithm: Employs AdaBoost as the core predictive engine.
• Validation Method: Uses the jackknife test to validate model performance.

Scientific Applications:

• Prediction Performance: Reported overall accuracies of 82.97% on BP1075 and 82.14% on BP594 with Cohen's kappa values of 0.66 and 0.64, respectively.
• Biological Research: Facilitates identification of DNA-binding proteins to inform studies of gene regulation, disease pathways, and potential therapeutic targets.

Methodology:

Feature extraction from primary sequences using Physicochemical Distance Transformation, PseAAC, and GSC-PseAAC; dimensionality reduction with UMAP combined with feature importance scoring; classification with AdaBoost; validation by jackknife test on BP1075 and BP594.