pSuc-FFSEA

pSuc-FFSEA predicts protein succinylation sites by fusing multiple sequence-derived and physicochemical feature encodings for classification.

Key Features:

• Feature Fusion Approach: Integrates multiple types of sequence information and physicochemical properties into a combined feature representation.
• Input Encodings: Employs EBGW (Extended Binary Grouped Words), One-Hot encoding, Continuous Bag-of-Words model, Chaos Game Representation, and AAF_DWT (Auto-Associative Flipping Discrete Wavelet Transform).
• LASSO Feature Selection: Uses the LASSO method to identify an optimal subset of features for classification.
• Stacking Ensemble Algorithm: Implements a two-layer stacking ensemble with SVM, Broad Learning System, and LightGBM as base classifiers and logistic regression as the meta-classifier.
• Hyperparameter Optimization: Applies Bayesian optimization and grid search for classifier hyperparameter tuning.

Scientific Applications:

• Protein Succinylation Prediction: Predicts protein succinylation sites and facilitates exploration of their roles in disease mechanisms and cellular regulation.

Methodology:

Feature fusion of EBGW, One-Hot, Continuous Bag-of-Words, Chaos Game Representation, and AAF_DWT followed by LASSO feature selection, a two-layer stacking ensemble with SVM, Broad Learning System and LightGBM as base classifiers and logistic regression as meta-classifier, and hyperparameter optimization via Bayesian optimization and grid search.