HemoNet

HemoNet predicts the hemolytic activity of peptide sequences by integrating SeqVec sequence embeddings with SMILES-based circular fingerprint representations of N/C-terminal modifications using a neural network.

Key Features:

• Neural network architecture: Uses a neural network to model relationships between sequence context and terminal chemical modification features.
• SeqVec contextual embeddings: Applies SeqVec embeddings to capture the contextual importance of amino acids within peptide sequences.
• SMILES-based circular fingerprint representation: Encodes N/C-terminal chemical modifications as SMILES-based circular fingerprint representations.
• Integrated sequence-chemistry modeling: Combines SeqVec embeddings with circular fingerprint representations to jointly model sequence context and terminal chemical structure.
• Performance: Reports an AUC-ROC of 88% and outperforms HemoPI and HemoPred (reported AUC-ROC of 73%).
• Validation strategies: Validated using stratified cross-validation, non-redundant cross-validation, and external peptide datasets including clinical antimicrobial peptides.

Scientific Applications:

• Therapeutic peptide discovery: Enables computational discovery and screening of novel therapeutic peptides by predicting hemolytic activity.
• Experimental prioritization: Prioritizes peptide candidates to guide wet-lab experimental screening based on predicted hemolytic profiles.
• Safety assessment in drug development: Supports safety and efficacy assessment of therapeutic agents by predicting hemolytic risk.
• Evaluation of clinical antimicrobial peptides: Assesses hemolytic properties of clinical antimicrobial peptides within external datasets.

Methodology:

Sequence contextualization via SeqVec embeddings; encoding of N/C-terminal modifications as SMILES-based circular fingerprints; neural network-based prediction; validation using stratified cross-validation, non-redundant cross-validation, and external peptide datasets including clinical antimicrobial peptides.