Development of software platform for the design of therapeutic antibodies

The first year of the PhD program focused on developing and validating a computational pipeline for designing specific antibodies. The pipeline consists of three main steps: computational modeling of the target antigen, protein-protein docking of the antibody-antigen complex to establish a starting point for affinity maturation, and the optimization of the antibody's binding affinity to the chosen target. In the absence of a detailed 3D model of the target protein, computational modeling was performed using AlphaFold2.2 (multimer version) based on the protein’s amino acid sequence. To validate this step, a dataset of 17 non-redundant TCRs was modeled using different software, including AlphaFold2, TCRModel, and TCRBuilder2. The best-performing method was AlphaFold2.2, achieving an 82.3% success rate with an RMSD below 2 Å. Once the target protein was modeled, the next step was protein-protein docking to generate an initial antibody-antigen complex. This step involved identifying antigenic epitopes and positioning the antibody accordingly. Docking was performed using the SnugDock protocol from Rosetta Suite, which optimizes binding energy by adjusting protein side chains. Validation was conducted using a dataset of 24 non-redundant antibody-antigen complexes. Two docking methods, standard and ensemble docking, were tested. Both achieved a 66.6% success rate, but the standard procedure produced higher-quality complexes with an interface RMSD between 3.5 and 10 Å. The final step, affinity maturation, was carried out using Rosetta Antibody Design (RAbD), a Monte Carlo-based algorithm that explores possible antibody conformations to optimize binding energy. Validation involved designing antibodies against a dataset of 24 complexes and analyzing the design risk ratio (DRR) to determine whether antibody modifications improved affinity. Binding energy calculations using MM/GBSA and MM/PBSA methods after molecular dynamics simulations indicated that the computationally optimized antibody complexes had lower binding energy than their experimental counterparts. Experimental validation is ongoing, with recombinant antibody production and binding affinity measurements in progress. Following the completion of the pipeline's development, work began on designing an antibody targeting T cells affected by Sézary syndrome, a cutaneous T-cell lymphoma. Collaboration with the Immaculata Dermopathic Institute provided access to the amino acid sequences of neoplastic TCRs from four patients. These sequences were used to computationally model the 3D structures of the neoplastic TCRs using AlphaFold2 (multimer version). Trastuzumab was selected as the antibody scaffold due to its high stability and tolerance for structural modifications. For each modeled TCR, a docking simulation with Trastuzumab was performed, focusing on selecting the best epitope for antibody binding. The β3 chain CDR of the TCR was chosen as the target due to its high structural variability. The best docking poses, as determined by SnugDock scoring, were selected for further optimization. Subsequently, 10,000 antibody-antigen complex structures were generated and evaluated based on their binding energy scores. Experimental validation focused on testing the computationally designed antibody against the Sézary patient with the highest percentage of neoplastic TCRs. Initial flow cytometry assays with patient cells did not yield conclusive results. Consequently, the TCR was expressed as a single-chain variable fragment (ScFv) to facilitate binding analysis via Surface Plasmon Resonance (SPR). SPR confirmed that the designed antibody successfully binds the neoplastic TCR with a dissociation constant in the nanomolar range. Future steps involve testing additional antibodies from the remaining patients and conducting further flow cytometry assays.