HemoPI

HemoPI predicts the hemolytic potential of peptides to assess their suitability for peptide therapeutic development.

Key Features:

• Dataset Generation: HemoPI-1 is a curated dataset comprising 552 experimentally validated hemolytic peptides from the Hemolytik database and an equal number of non-hemolytic peptides from Swiss-Prot.
• Sequence Analysis: Identification of residues (Leucine [L], Lysine [K], Phenylalanine [F], Tryptophan [W]) and motifs ("FKK", "LKL", "KKLL", "KWK", "VLK", "CYCR", "CRR", "RFC", "RRR", "LKKL") that are prevalent in hemolytic peptides.
• Machine Learning Models: Models trained using machine learning techniques to distinguish hemolytic and non-hemolytic peptides with reported accuracy exceeding 95%.
• Potential Discrimination: Additional models (HemoPI-2 and HemoPI-3) that classify peptides by hemolytic potential into high or low potency groups.

Scientific Applications:

• Virtual Screening: Prioritizing peptide candidates by predicted hemolytic potential during in silico screening campaigns.
• Analog-Based Peptide Design: Guiding sequence modifications to reduce hemolytic activity based on residue and motif associations.
• Early-Stage Safety Assessment: Assessing hemolytic risk of peptide leads to reduce attrition due to hemolytic activity in preclinical or clinical stages.

Methodology:

Curated HemoPI-1 dataset from Hemolytik and Swiss-Prot; sequence analysis to identify specific residues and motifs associated with hemolysis; and training of machine learning models to classify hemolytic versus non-hemolytic peptides and to discriminate hemolytic potency (HemoPI-2, HemoPI-3) with reported accuracy >95%.