StarPepDB

StarPepDB maps and analyzes the chemical space of antimicrobial and antiparasitic peptides (AMPs and APPs) using network-science representations and group-fusion similarity-based searches to identify and prioritize candidate antiparasitic peptides.

Key Features:

• Network-Based Representation: Represents the antiparasitic peptide chemical space using three network types: Chemical Space Network (CSN), Half-Space Proximal Network (HSPN), and Metadata Network (METN).
• Centrality Measures: Computes centrality measures to identify the most significant and nonredundant peptides within each network for use as queries.
• Group-Fusion Similarity-Based Searches: Uses central peptides as queries in group-fusion, multiquery similarity-based searches against the StarPepDB AMP collection.
• Multiquery Similarity-Based Search Models (mQSSMs): Builds mQSSMs from fused multiquery results to propose new potential APPs.
• Performance Evaluation: Evaluates mQSSMs on five benchmarking APP/non-APP datasets reporting Matthews correlation coefficient (MCC) values from 0.834 to 0.965, with a model using 219 queries from CSN and HSPN achieving MCC > 0.85.
• Comparison with Existing Models: Demonstrates superior predictive accuracy relative to APP prediction servers AMPDiscover and AMPFun.
• Repurposing and Lead Identification: Applies top mQSSM models and filtering to repurpose 95 AMPs as potential APP hits and to identify 11 APP lead candidates via diversity-based network analyses.
• Motif Identification Approaches: Employs computational approaches to identify relevant sequence motifs for searching and designing new APPs.
• Integration with Analysis Servers: Integrates results with 24 web servers for assessing activity, toxicity, and drug-like properties.

Scientific Applications:

• APP discovery: Identify and prioritize novel antiparasitic peptide candidates from the AMP chemical space.
• AMP repurposing: Reclassify existing AMPs as potential APPs through multiquery similarity models and filtering.
• Lead selection: Select diverse APP lead candidates using centrality-informed queries and network diversity analyses.
• Motif discovery and peptide design: Extract sequence motifs to guide APP searching and rational peptide design.
• Benchmarking and model validation: Provide quantitative evaluation of predictive models on curated APP/non-APP benchmarking datasets using MCC.

Methodology:

Constructs CSN, HSPN and METN networks; applies centrality measures to select representative nonredundant peptides; performs group-fusion multiquery similarity searches to build mQSSMs against the StarPepDB collection; evaluates models on five benchmarking APP/non-APP datasets (MCC 0.834–0.965); and uses computational motif-identification approaches plus filtering and diversity-based network analyses to repurpose AMPs and select lead candidates.