Machine Learning Applied to Determine the Molecular Descriptors Responsible for the Viscosity Behavior of Concentrated Therapeutic Antibodies.

Predicting the solution viscosity of monoclonal antibody (mAb) drug products remains as one of the main challenges in antibody drug design, manufacturing, and delivery. In this work, the concentration-dependent solution viscosity of 27 FDA-approved mAbs was measured at pH 6.0 in 10 mM histidine-HCl. Six mAbs exhibited high viscosity (>30 cP) in solutions at 150 mg/mL mAb concentration. Combining molecular modeling and machine learning feature selection, we found that the net charge in the mAbs and the amino acid composition in the Fv region are key features which govern the viscosity behavior. For mAbs whose behavior was not dominated by charge effects, we observed that high viscosity is correlated with more hydrophilic and fewer hydrophobic residues in the Fv region. A predictive model based on the net charges of mAbs and a high viscosity index is presented as a fast screening tool for classifying low- and high-viscosity mAbs.

Machine Learning Models of Antibody-Excipient Preferential Interactions for Use in Computational Formulation Design.

Preferential interactions of formulation excipients govern their impact on the stability properties of proteins in solution. The ability to predict these interactions without the need to perform experiments would enable formulation design to begin early in the development of a new antibody therapeutic. With that in mind, we developed a feature set to numerically describe local regions of an antibody's surface for use in machine learning applications. Then, we used these features to train machine learning models for local antibody-excipient preferential interactions for the excipients sorbitol, sucrose, trehalose, proline, arginine·HCl, and NaCl. Our models had accuracies of up to about 85%. We also used linear (elastic net) models to quantify the contribution of antibody surface features to the preferential interaction coefficients, finding that the carbohydrates and proline tend to have similar important features, while the interactions of arginine·HCl and NaCl are governed by charge features. We present several case studies demonstrating how these machine learning models could be used to predict experimental aggregation and viscosity behavior in solution. Finally, we propose an approach to computational formulation design wherein a panel of excipients may be considered while designing an antibody sequence.

Machine Learning Feature Selection for Predicting High Concentration Therapeutic Antibody Aggregation.

Protein aggregation can hinder the development, safety and efficacy of therapeutic antibody-based drugs. Developing a predictive model that evaluates aggregation behaviors during early stage development is therefore desirable. Machine learning is a widely used tool to train models that predict data with different attributes. However, most machine learning techniques require more data than is typically available in antibody development. In this work, we describe a rational feature selection framework to develop accurate models with a small number of features. We applied this framework to predict aggregation behaviors of 21 approved monospecific monoclonal antibodies at high concentration (150 mg/mL), yielding a correlation coefficient of 0.71 on validation tests with only two features using a linear model. The nearest neighbors and support vector regression models further improved the performance, which have correlation coefficients of 0.86 and 0.80, respectively. This framework can be extended to train other models that predict different physical properties.

Molecular computations of preferential interactions of proline, arginine.HCl, and NaCl with IgG1 antibodies and their impact on aggregation and viscosity.

Preferential interactions of excipients with the antibody surface govern their effect on the stability of antibodies in solution. We probed the preferential interactions of proline, arginine.HCl (Arg.HCl), and NaCl with three therapeutically relevant IgG1 antibodies via experiment and simulation. With simulations, we examined how excipients interacted with different types of surface patches in the variable region (Fv). For example, proline interacted most strongly with aromatic surfaces, Arg.HCl was included near negative residues, and NaCl was excluded from negative residues and certain hydrophobic regions. The differences in interaction of different excipients with the same surface patch on an antibody may be responsible for variations in the antibody's aggregation, viscosity, and self-association behaviors in each excipient. Proline reduced self-association for all three antibodies and reduced aggregation for the antibody with an association-limited aggregation mechanism. The effects of Arg.HCl and NaCl on aggregation and viscosity were highly dependent on the surface charge distribution and the extent of exclusion from highly hydrophobic patches. At pH 5.5, both tended to increase the aggregation of an antibody with a strongly positive charge on the Fv, while only NaCl reduced the aggregation of the antibody with a large negative charge patch on the Fv. Arg.HCl reduced the viscosities of antibodies with either a hydrophobicity-driven mechanism or a charge-driven mechanism. Analysis of this data presents a framework for understanding how amino acid and ionic excipients interact with different protein surfaces, and how these interactions translate to the observed stability behavior.