High-throughput self-interaction chromatography: applications in protein formulation prediction.

Area of Science:

• Biophysical Chemistry
• Computational Biology
• Protein Formulation Science

Background:

• Understanding protein self-interactions is crucial for developing stable formulations.
• The second virial coefficient (B22) quantifies protein-protein interactions and is linked to solubility.
• Predicting B22 across diverse formulation conditions is challenging but essential for efficient formulation development.

Purpose of the Study:

• To demonstrate the predictive capability of an artificial neural network (ANN) for protein self-interactions.
• To train an ANN on measured second virial coefficient (B22) data from a limited formulation screen.
• To predict B22 values for untested formulation conditions, optimizing protein solubility.

Main Methods:

• Measured B22 values for lysozyme across 81 formulation conditions using self-interaction chromatography (SIC).
• Developed and trained an ANN model using the measured B22 data and formulation parameters (pH, salt, additives).
• Validated ANN predictions by experimentally measuring B22 for 20 selected untested conditions.

Main Results:

• The ANN accurately predicted lysozyme B22 values with high statistical significance (p<0.0001).
• Achieved a Root Mean Square Error (RMSE) of 2.6x10⁻⁴ mol ml/g², indicating precise predictions.
• Experimental validation confirmed the ANN's predictive accuracy for protein self-interactions.

Conclusions:

• An ANN trained on a small dataset of B22 measurements can reliably predict protein self-interactions.
• This approach enables accurate prediction of B22 values for numerous untested formulation conditions.
• Predictive modeling accelerates the identification of formulations that enhance protein solubility and stability.