Proteins regulate numerous biological processes, and their physicochemical properties determine their stability, structural flexibility, and interactions with other molecules. For formulations containing therapeutic proteins such as monoclonal antibodies, it is therefore essential to test chemical degradation and aggregation of the active substance, since these phenomena affect the drug’s safety and efficacy. A critical factor for initiating clinical trials is also the initial dose estimate, which is strongly dependent on the drug’s bioavailability. Approaches which would be capable of reliably predicting whether a given monoclonal antibody will be suitable for further development already at the molecule-design stage are becoming increasingly important.
In this master's thesis, we investigate the use of mathematical–physical modelling in combination with experiments to predict protein properties, specifically those of monoclonal antibodies, using surface descriptors. To achieve this, we employed AlphaFold for protein structure prediction and molecular dynamics simulations to analyze their surface properties. Various descriptors, including SASA (solvent accessible surface area), SCM (spatial charge map), and others, were calculated and used in modeling bioavailability.
Despite a systematic approach and a broad set of descriptors, results indicate that the selected parameters are insufficient for reliably predicting bioavailability, highlighting the complexity of the processes involved and the need for further methodological improvements. As an example of a successful application of surface descriptors for protein property prediction, we also present an accurate prediction of protein aggregation. Our findings emphasize the limitations of our chosen approach and raise new research questions about improving predictive models for understanding protein properties.
|
