QSAR model development for early stage screening of monoclonal antibody therapeutics to facilitate rapid developability

Abstract

Monoclonal antibodies (mAbs) and related therapeutics are highly desirable from a biopharmaceutical perspective as they are highly target specific and well tolerated within the human system. Nevertheless, several mAbs have been discontinued or withdrawn based either on their inability to demonstrate efficacy and/or due to adverse effects. With nearly 80% of drugs failing in clinical development mainly due to lack of efficacy and safety there arises an urgent need for better understanding of biological activity, affinity, pharmacology, toxicity, immunogenicity etc. thus leading to early prediction of success/failure. In this study a hybrid modelling framework was developed that enabled early stage screening of mAbs. The applicability of the experimental methods was first tested on chemical compounds to assess the assay quality following which they were used to assess potential off target adverse effects of mAbs. Furthermore, hypersensitivity reactions were assessed using Skimune™, a non-artificial human skin explants based assay for safety and efficacy assessment of novel compounds and drugs, developed by Alcyomics Ltd. The suitability of Skimune™ for assessing the immune related adverse effects of aggregated mAbs was studied where aggregation was induced using a heat stress protocol. The aggregates were characterised by protein analysis techniques such as analytical ultra-centrifugation following which the immunogenicity tested using Skimune™ assay. Numerical features (descriptors) of mAbs were identified and generated using ProtDCal, EMBOSS Pepstat software as well as amino acid scales for different. Five independent and novel X block datasets consisting of these descriptors were generated based on the physicochemical, electronic, thermodynamic, electronic and topological properties of amino acids: Domain, Window, Substructure, Single Amino Acid, and Running Sum. This study describes the development of a hybrid QSAR based model with a structured workflow and clear evaluation metrics, with several optimisation steps, that could be beneficial for broader and more generic PLS modelling. Based on the results and observation from this study, it was demonstrated incremental improvement via selection of datasets and variables help in further optimisation of these hybrid models. Furthermore, using hypersensitivity and cross reactivity as responses and physicochemical characteristics of mAbs as descriptors, the QSAR models generated for different applicability domains allow for rapid early stage screening and developability. These models were validated with external test set comprising of proprietary compounds from industrial partners, thus paving way for enhanced developability that tackles manufacturing failures as well as attrition rates.