Hybrid peptide classifier model for predicting periodontal cell-penetrating peptides

Background: Periodontitis, a global infectious disease-causing tooth loss and tissue destruction, is linked to diabetes and cardiovascular diseases. Conventional treatments fail to eliminate pathogens, necessitating alternative therapies. Cell-penetrating peptides (CPP) are promising for therapeutic applications like genetic defect correction and gene silencing, but face challenges like cytotoxicity and immune responses. They also manage periodontal disease by delivering agents directly to targeted tissues, improving drug penetration and treatment outcomes. CPP unique ability to traverse cellular membranes is key. Hybrid Peptide Classifier, a novel model using an LLM-based attention network, combines the strengths of multiple neural network layers to model peptide sequence structure and dependencies effectively. By improving medication delivery straight to infected periodontal sites, CPP provide a novel treatment option for periodontitis because of their antimicrobial activity and tissue-penetrating capacity. This model aims to predict periodontal cell-penetrating peptides, accelerating advancements in peptide-based therapies and drug delivery systems. Methods: The peptide classification dataset was sourced from thegleelab.org/MLCPP/MLCPPData.html, featuring sequences for both positive and negative sample classes. A custom PyTorch Dataset class was related to maintaining a consistent sequence length. The dataset was split into training and testing subsets and loaded into DataLoader objects for efficient batch processing. The hybrid peptide classifier class is a neural network model designed for peptide classification, initialized with vocabulary size, embedding dimension, hidden dimension, and maximum sequence length, and subjected to training with an epoch of 10 with early stopping. A hybrid architecture comprising convolutional and bidirectional LSTM layers was used to categorize peptide sequences. Results: The model exhibited strong classification performance with an accuracy of 85.2%, an F1-score of 0.88, and an AUC of 0.93. Conclusion: CPPs are promising tools for drug delivery and gene therapy, but challenges like data imbalances and experimental variability must be addressed. Our study showed promising results in better classifying the peptide sequences. Future research should focus on refining machine learning techniques, integrating diverse datasets, and implementing rigorous validation protocols to improve peptide classification models' reliability and patient outcomes in peptide-based therapeutics. This model provides a basis for creating customized, targeted peptide treatments in periodontology.