CancerGram

CancerGram predicts anticancer peptides (ACPs) using n-gram sequence encoding and random forest classification to identify cationic peptides that selectively target cancer cell mitochondrial and plasma membranes.

Key Features:

• n-gram encoding: Represents peptide sequences using n-gram features extracted from amino acid sequences.
• Random forest classification: Employs random forest models as the primary machine learning classifier.
• Three-class model: Classifies peptides into anticancer peptides (ACPs), antimicrobial peptides (AMPs), and non-anticancer/non-antimicrobial peptides.
• Performance metrics: Achieves reported AU1U of 0.89 and Kappa statistic of 0.65 for the three-class model.
• Peptide physicochemical characterization: Accounts for peptide properties such as small size, positive charge, hydrophobicity, and amphipathicity.
• Membrane interaction focus: Targets interactions with negatively charged components of biological membranes and specifically mitochondrial and plasma membranes of cancer cells.

Scientific Applications:

• Candidate screening: Prioritizes cationic peptide candidates for experimental testing as potential anticancer agents.
• Discrimination of peptide function: Distinguishes ACPs from AMPs and non-ACP/non-AMP peptides to guide biological validation.
• Therapeutic discovery: Facilitates computational discovery of peptide-based therapeutics targeting cancer cell membranes.
• Cost reduction in experimental workflows: Reduces the experimental search space by computationally prioritizing likely ACPs.

Methodology:

Uses n-gram sequence encoding and random forest classification to build a three-class predictive model (ACPs, AMPs, non-anticancer/non-antimicrobial peptides) with reported AU1U = 0.89 and Kappa = 0.65.