PITHA

PITHA predicts immunogenicity of humanized and fully human therapeutic antibodies by analyzing sequence and co-crystal structural features and applying machine-learning classification.

Key Features:

• Integration of Sequence and Structural Data: Uses sequence information and co-crystal structures of idiotypic and anti-idiotypic antibodies to characterize binding interactions that predominantly occur at complementarity-determining regions (CDRs).
• Identification of Immunogenic Features: Extracts and evaluates specific features associated with immunogenicity, notably cavity volume at the CDR region and hydrophobicity within the CDR-H3 loop.
• Machine Learning Integration: Incorporates extracted features into a machine-learning classifier and reports an accuracy of 83% in leave-one-out cross-validation on 29 therapeutic antibodies with available crystal structures.

Scientific Applications:

• Computer-aided antibody design: Predicts potential immunogenicity of new therapeutic antibodies to inform candidate selection and modification.
• Early immunogenicity assessment: Assesses likelihood of anti-idiotypic antibody responses mediated by CDR regions to guide safety optimization.
• Prioritization of candidates: Ranks therapeutic antibodies for further development based on predicted immunogenic potential to support resource allocation.

Methodology:

Uses structural data from co-crystal antibody–anti-idiotype complexes for data acquisition; extracts features including CDR cavity volume and CDR-H3 hydrophobicity; trains a machine-learning model on known therapeutic antibody data and evaluates performance using leave-one-out cross-validation on 29 antibodies with crystal structures (83% accuracy).