A DEEP EVOLUTIONARY WORD EMBEDDING FRAMEWORK FOR ROBUST AMYLOID PROTEIN CLASSIFICATION

Abstract

Amyloid proteins are central to the pathogenesis of numerous neurodegenerative and systemic disorders, including Alzheimer's disease, Parkinson's disease, and amyloidosis-related conditions. Accurate computational identification of amyloidogenic proteins is therefore essential for early diagnosis and therapeutic research. This study proposes a deep evolutionary word embedding framework for robust classification of amyloid and non-amyloid proteins. The approach integrates evolutionary information derived from protein sequences with advanced word embedding techniques to capture both local and global contextual patterns. Protein sequences are first transformed into overlapping k-mers and enriched with evolutionary profiles derived from Position-Specific Scoring Matrices (PSSMs), enabling the model to preserve evolutionary conservation and mutation tendencies. These enriched representations are then mapped into continuous vector spaces using a deep embedding model, effectively learning discriminative semantic features from biological sequences. A hybrid deep learning classifier combining convolutional and recurrent layers is employed to exploit spatial and sequential dependencies within the embedded representations. Extensive experiments were conducted on benchmark amyloid protein datasets using five-fold cross-validation, achieving an average accuracy of 93.4%, precision of 0.94, recall of 0.93, F1-score of 0.93, Matthews correlation coefficient of 0.86, and AUC of 0.96. The proposed framework consistently outperforms conventional machine learning techniques and shallow feature-based approaches across all evaluation metrics. These results confirm that incorporating evolutionary-aware deep word embeddings substantially enhances the predictive capability of amyloid protein identification models. This framework offers a powerful and flexible computational tool to support large-scale proteomic analysis and advance research on amyloid-related diseases and drug discovery.

Keywords: Amyloid Protein Classification, Evolutionary Word Embedding, Protein Sequence Analysis, Deep Learning, Machine Learning, Bioinformatics