Rational design and data-driven strategies to optimize bioconjugation processes in antibody-drug conjugates

The main focus of this thesis is to address the challenges in developing efficient and cost-effective drug delivery systems. Among the most promising approaches are antibody-drug conjugates (ADCs), which combine cytotoxic or bioactive agents with monoclonal antibodies (mAbs) to achieve targeted therapies. However, bioconjugation processes can lead to variable outcomes depending on the choice of mAb, amino acid residues, and linker-payload (LP) systems, making the design of effective ADCs a complex task. In this work, a machine learning (ML) framework capable of predicting bioconjugation outcomes is presented, thereby guiding the selection of optimal mAb-LP combinations and reaction conditions. Specifically, the eXtremeGradientBoosting (XGBoost) algorithm is used to model and predict the drug-to-antibody ratio (DAR) in ADC synthesis. The proposed approach demonstrates high predictive accuracy, achieving R2 scores of 0.85 and 0.91 for lysine- and cysteine- conjugated datasets, respectively. By integrating ML algorithms into the design and optimization of ADC bioconjugation processes, this study provides a data-driven strategy to streamline ADC development and improve the efficiency of targeted drug delivery systems. In addition, during my period at The Scripps Research Institute, LaJolla (CA). I contributed to a project aimed at improving the existing reactive docking methodology, designed to model and predict reactions between small molecules and biological macromolecules. In this work, pseudo-atoms (PAs) were introduced on the ligand warhead to encode the geometry and spatial orientation necessary for covalent bond formation, enabling the prediction of the optimal near-attack conformation (NAC). Here, I present the preliminary results obtained from reactive docking using PAs on two cysteine-reactive warheads (chloroacetamides and acrylamides), in predicting the correct reactive residues and the optimal geometric approach for covalent bond formation.