Neural Networks for Predicting and Classifying Antimicrobial Resistance Sequences in Porphyromonas gingivalis

Porphyromonas gingivalis is a key pathogen associated with periodontal disease linked to various systemic conditions. Accurate identification of P. gingivalis proteins is essential for understanding its pathogenicity and developing targeted interventions. Recent advances in whole-genome sequencing of P. gingivalis have enhanced the detection and classification of antimicrobial resistance (AMR) determinants, aiding in the early identification of resistance trends and improving patient care. In this study, we developed a deep learning approach using convolutional neural networks (CNNs) to classify P. gingivalis proteins based on their amino acid sequences.

A dataset of 685 protein sequences, including 150 P. gingivalis proteins and 535 nonresistant variants, was compiled and split into training (60%), validation (20%), and test (20%) sets. The sequences were preprocessed by padding to 750 amino acids and one-hot encoded into a feature matrix. A CNN model, consisting of two convolutional layers, max pooling, dropout, and fully connected layers for binary classification, was designed and implemented in PyTorch with 6192,258 parameters. The model was trained using the Adam optimizer for 30 epochs with early stopping based on validation accuracy.

The CNN model outperforms traditional methods like BLAST, HMM Profiles, and DeepSig in predicting and classifying AMR in P. gingivalis. The hypothetical ProtBERT model shows slightly better performance, with an accuracy of 97%. Key metrics like accuracy, precision, recall, F1 score, and the area under the curve were assessed. CNN and ProtBERT have high recall rates (0.93 and 0.95, respectively), indicating their effectiveness in predicting AMR classifications.

Our CNN model outperforms SOTA methods in classifying P. gingivalis-resistant protein sequences, achieving 96.35% accuracy and an area under the curve of 0.98.

Precise and rapid prediction of AMR based solely on protein sequences, potentially leading to earlier identification of resistance trends and improved antibiotic stewardship in periodontal treatment.