Synergistic Ensemble Methods for Predicting VEGF Sequences Associated with Dental Cystic Lesions and Ameloblastomas

Abstract

Introduction: VEGF/VPF is a 45kd glycoprotein found on human chromosome 6p21.3 that binds to endothelial cell receptors. Five forms of the VEGF gene have been identified in mammals, promoting endothelial cell proliferation and migration and contributing to normal angiogenesis and pathological conditions like tumorigenesis. Targeting VEGF receptors is crucial for treating dentigerous cysts and tumors, requiring a sequence-based understanding for targeted therapies and surgical success. The study explores the role of VEGF in dental cystic lesions and Ameloblastomas, aiming to improve clinical outcomes by understanding tumor behavior. It uses ensemble methods for predicting VEGF sequences, integrating multi-omics data and machine learning to provide insights into molecular pathology and improve patient outcomes.
Methods: The human VEGF sequences, including P15692, A0A0A0MTB2, |P49765, O43915, A0A0A0MSI7, A0A0A0MRQ4, and Q7LAP4, were retrieved, checked for missing values, and analyzed for prediction using Protbert embeddings and stacked ensemble learning. The study used a dataset of 1,034 features from advanced measurements to enhance signal-to-noise ratio, reduce dimensionality, and retain critical information. The dataset was preprocessed using feature scaling and a hybrid feature selection method. The Standard Scaler standardized features, while the XGBoost algorithm refined selection. The final set of 459 optimized features was combined through stacking.
Results: The stacked ensemble model achieved 70% accuracy, relying heavily on gradient-boosting techniques. XGBoost optimizes memory usage and handles missing data. Random Forest may not effectively extract dataset complexities. Classifier weight analysis guides future strategies, focusing on boosting methods.
Conclusion: The study developed an ensemble model to predict VEGF-related outcomes in ameloblastoma and dentigerous cysts patients, but challenges like data imbalance and overfitting need to be addressed.