antifp

antifp predicts and designs antifungal peptides (AFPs) using support vector machine-based classifiers trained on a dataset of 1459 known antifungal peptides to identify sequence features associated with antifungal activity.

Key Features:

• Dataset: Uses a curated dataset of 1459 known antifungal peptides (AFPs) for model training and evaluation.
• Predictive models: Employs support vector machine (SVM) classifiers built on multiple peptide feature representations.
• Feature types: Models use residue composition, binary profiles, terminal-residue patterns, and positional preference information as input features.
• Residue analysis: Reports higher abundance of Cysteine (C), Glycine (G), Histidine (H), Lysine (K), Arginine (R), and Tyrosine (Y), with positional preferences such as Arginine (R), Valine (V), and Lysine (K) at the N-terminus and Cysteine (C) and Histidine (H) at the C-terminus.
• Model accuracy (compositional SVM): Achieves 88.78% accuracy on the training dataset and 83.33% on an independent validation dataset using compositional features.
• Model accuracy (binary terminal patterns): Binary-pattern models of terminal residues achieve 84.88% accuracy on training and 84.64% on validation datasets.
• Benchmarking: Models were benchmarked against existing methods using a dataset of compositionally similar antifungal and non-AFPs, with the binary-based model outperforming other methods.

Scientific Applications:

• Discovery of novel antifungal agents: Prediction of candidate AFP sequences for experimental follow-up and therapeutic development.
• Peptide design: Design and optimization of novel peptides with predicted antifungal activity based on compositional and terminal-residue features.
• Understanding peptide functionality: Identification of sequence and positional residue patterns correlated with antifungal activity to inform structure–function studies.

Methodology:

Machine-learning using support vector machines (SVM) trained on features including residue composition, binary profiles of terminal residues, and positional preference data, with benchmarking against compositionally similar antifungal and non-AFP datasets.