Sequence-Based Viscosity Prediction for Rapid Antibody Engineering.

Area of Science:

• Biotechnology
• Protein Engineering
• Computational Biology

Background:

• High protein concentration formulations are essential for effective therapeutic delivery.
• Antibody aggregation and high viscosity can limit therapeutic efficacy and administration routes.
• Predictive models are needed to guide protein engineering efforts for improved formulation properties.

Purpose of the Study:

• To develop a machine learning model for predicting antibody viscosity based on amino acid sequence.
• To apply a structure-based rational design strategy combined with in silico screening to reduce antibody viscosity.
• To efficiently engineer a highly viscous anti-IL-13 monoclonal antibody candidate.

Main Methods:

• Machine learning algorithms were employed to identify correlations between antibody amino acid sequences and their physical characteristics.
• A structure-based rational design approach was used to generate antibody variants with predicted lower viscosity.
• An in silico viscosity prediction tool was utilized to screen and prioritize engineered variants before experimental validation.

Main Results:

• A predictive model for antibody internal viscosity was successfully developed.
• A panel of 16 variants of a highly viscous anti-IL-13 monoclonal antibody was generated.
• The combined strategy efficiently reduced the viscosity of the anti-IL-13 antibody candidate from 34 cP to 13 cP at 150 mg/mL.

Conclusions:

• Integrating machine learning-based viscosity prediction with rational design accelerates the engineering of low-viscosity antibody therapeutics.
• This approach enables efficient optimization of antibody candidates for high-concentration formulations.
• The developed methodology offers a powerful tool for improving the developability of therapeutic antibodies.