Predictive QSAR tools to aid in early process development of monoclonal antibodies

Abstract

Monoclonal antibodies (mAbs) have become one of the fastest growing markets for diagnostic and therapeutic treatments over the last 30 years with a global sales revenue around $89 billion reported in 2017. A popular framework widely used in pharmaceutical industries for designing manufacturing processes for mAbs is Quality by Design (QbD) due to providing a structured and systematic approach in investigation and screening process parameters that might influence the product quality. However, due to the large number of product quality attributes (CQAs) and process parameters that exist in an mAb process platform, extensive investigation is needed to characterise their impact on the product quality which makes the process development costly and time consuming. There is thus an urgent need for methods and tools that can be used for early risk-based selection of critical product properties and process factors to reduce the number of potential factors that have to be investigated, thereby aiding in speeding up the process development and reduce costs. In this study, a framework for predictive model development based on Quantitative Structure-Activity Relationship (QSAR) modelling was developed to link structural features and properties of mAbs to Hydrophobic Interaction Chromatography (HIC) retention times and expressed mAb yield from HEK cells. Model development was based on a structured approach for incremental model refinement and evaluation that aided in increasing model performance until becoming acceptable in accordance to the OECD guidelines for QSAR models. The resulting models showed that it was possible to predict HIC retention times of mAbs based on their inherent structure. Further improvements of the models are suggested due to performance being adequate but not sufficient for implementation as a risk assessment tool in QbD. However, the described methodology and workflow has been proven to work for retention time prediction in a HIC column and is therefore likely to be applicable to other purification columns.