Machine learning predictions of IgG1 and IgG4 self-association and high-concentration solution properties

Area of Science:

• Biopharmaceutical Development
• Protein Aggregation Science
• Antibody Engineering

Background:

• Subcutaneous delivery of antibody therapeutics requires molecules with low viscosity and aggregation propensity in concentrated formulations.
• Early identification of antibodies with low self-association improves late-stage development success.

Purpose of the Study:

• To experimentally profile the self-association behavior of clinical-stage antibodies.
• To identify formulation conditions that suppress antibody self-association.
• To develop predictive models for antibody self-association and high-concentration properties.

Main Methods:

• Experimental profiling of antibody self-association across various pH and excipient conditions.
• Correlation analysis between self-association, viscosity, and recovery after concentration.
• Development and validation of machine learning models for predicting self-association.

Main Results:

• Acidic formulations (pH 5) with 200 mM proline effectively suppress self-association in both IgG1 and IgG4 antibodies.
• Self-association measurements correlate with viscosity and are inversely related to recovery after concentration.
• Machine learning models accurately predict IgG1 and IgG4 self-association, identifying favorable high-concentration properties.

Conclusions:

• Formulation strategies, particularly acidic pH with proline, can mitigate antibody self-association.
• Predictive models enhance the identification of drug-like antibody molecules early in development.
• These findings streamline the antibody development process for subcutaneous delivery.