Comparative Optimization of Sequence-Based Feature Encodings for Antimicrobial Peptide Prediction Using Random Forest on Deep-AmPEP30

Antimicrobial peptides, also called Host Defense Peptides (HDPs) are short-length protein sequences. Antimicrobial peptides are effective against antibioticresistant pathogens. Antibiotic resistance in pathogens is a future threat to human life caused by the overuse of antibiotics. This research aims to compare and identify the most efficient encoding technique to accurately detect antimicrobial peptides. In this research, performance comparison is evaluated using different feature extraction techniques performed on the DeepAmPEP30 dataset. The evaluated techniques include amino acid composition (AAC), dipeptide amino acid composition (DPC), pseudo amino acid composition (PseAAC), pseudo k-tuple reduced amino acid composition (PseKRAAC), reduced amino acid composition (RAAC), composition transition and distribution (CTD), and k-mer encoding techniques. These extracted features serve as input for the random forest classifier model, and performance is assessed and compared. The research findings reveal that the CTD shows the highest accuracy of 87%. High precision, recall and F1 score values are also shown by this encoding scheme when compared to others. Five-fold cross validation and paired t-test are also conducted to prove the generalizability and significance of the result obtained. The research also provides valuable insights into diverse feature representation methods crucial for identifying complex patterns in antimicrobial peptides (AMPs) predictions, offering a foundation for discovering novel antimicrobial peptides to address antimicrobial resistance effectively.