We heartly congratulate our colleague Abir Omran on obtaining her Ph.D. in the field of Pharmacy! Abir graduated from the master programme in pharmaceutical modelling from Uppsala University. As PhD student in the Pharmacoinformatics Research Group she worked on toxicity of biological drugs, such as monoclonal antibodies. Furthermore, she was a PhD Student in the MolTag Doctoral Program. The main aim of her doctoral thesis, based on a cooperation with Sanofi, was to predict toxicity of biological drugs.        

Thesis Abstract

Biological drugs are large and complex molecules that have successfully treated diseases once considered challenging. With the continued growth of biological drugs, the need to understand their side effects and to develop computational tools to better assess their toxicity is increasing. The main aim of this thesis was to explore the toxicity prediction of biological drugs by utilizing data from available databases, analyzing the data, and investigating different ways to describe these macromolecules for machine learning. The thesis consists of three studies, all of which focus on the toxicity of biological drugs. Study 1 presents an interactive dashboard that utilizes side effect data based on Preferred Terms (PTs) from the MedDRA terminology and maps it to the other hierarchical terms. This enables the user to analyze the data across four different levels for preclinical, clinical, and post-marketing data. Study 2 investigates the side effect landscape based on side effects reported as adverse events (AEs) and adverse drug reactions (ADRs) for therapeutic mAbs. Additionally, Study 2 focused on neutropenia as a use case, highlighting the lack of concordance between animal models and human outcomes in predicting neutropenia. The findings also showed that, by simply using the therapeutical information of the drug, neutropenia reported as an AE was more predictable than when using animal data. The final study, Study 3, explores the prediction of immunogenicity by testing different representations and machine learning methods. The result showed that the simplest representation format, combined with a random forest algorithm, resulted in the best overall performance.

Keywords

Adverse Events / Adverse Drug Reactions / Therapeutic monoclonal antibodies / Biological drugs / Immunogenicity