Divergent Approaches to Antimicrobial Peptide Classification: A Comparative Evaluation of AI Techniques

Given the global rise in antibiotic resistance, antimicrobial peptides (AMPs) provide a viable substitute for traditional antibiotics. Through a comparison of transformer-based, deep learning (DL), and machine learning (ML) techniques, this work seeks to classify AMPs. AMP sequences from curated databases were compared with non-AMP sequences obtained from UniProtKB/Swiss-Prot. The study compares performance across many assessment variables to assess how well various algorithms separate AMPs from non-AMPs. Significantly, RoBERTa outperformed DistilBERT (84.43%) in accuracy (86.53%). To provide a thorough knowledge of the algorithms’ performance, the research also included a variety of other assessment criteria, including area under the receiver operating characteristic curve (AUC-ROC), recall, F1 score, and accuracy. The performance of the LSTM and GRU models was notable, despite their significantly lower accuracies. Furthermore, conventional machine learning methods such as Naive Bayes and Logistic Regression demonstrated competitive accuracy. This thorough analysis clarifies the benefits and drawbacks of various AMP classification techniques, providing insightful information for next studies on antimicrobial peptides and their potential to fight antibiotic resistance.