HLPpred-Fuse

HLPpred-Fuse predicts hemolytic peptides (HLPs) and classifies their hemolytic activity level to support assessment of peptide toxicity in drug development and immunoinformatics.

Key Features:

• Two-Layer Prediction Framework: Implements a two-layer framework that first discriminates HLPs from non-HLPs and then classifies activity level as high or low.
• Comprehensive Feature Representation: Generates 54 probabilistic features by integrating nine sequence-based encodings: amino acid composition (AAC), dipeptide composition (DPC), amino acid index (AAI), binary profile (BPF), composition-transition-distribution (CTD), conjoint triad (CTF), quasi-sequence order (QSO), grouped dipeptide composition (GDPC), and grouped tripeptide composition (GTPC).
• Machine Learning Integration: Uses six classifiers—Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting (GB), Extremely Randomized Trees (ERT), k-Nearest Neighbors (KNN), and AdaBoost (AB)—to produce probabilistic features.
• Feature Fusion: Fuses the 54 probabilistic features into a converged sequence representation and applies an Extremely Randomized Trees (ERT) algorithm to build two final prediction models for HLP identification and activity classification.

Scientific Applications:

• Drug Development: Assesses hemolytic potential of peptides to inform safety evaluation of peptide-based therapeutics.
• Immunoinformatics Research: Supports analysis of sequence features associated with hemolytic activity for studies of peptide structure–function relationships.

Methodology:

Feature generation: nine sequence-based encodings combined with six machine learning classifiers produce 54 probabilistic features per peptide; feature fusion and prediction: these features are fused into a converged sequence representation that is input to an Extremely Randomized Trees (ERT)-based model to develop two separate predictors for HLP identification and activity level.