Antibody Modeling Assessment Demonstrates the Reliability of Automated Methods
Antibodies are increasingly important in medical diagnostics and in the treatment of a broad range of disease states including cancer, inflammation and auto-immune diseases. Through antibody drug conjugates (ADCs), antibodies also enable the targeted delivery of traditional drugs. In contrast to traditional chemotherapeutic agents, ADCs target and attack the cancer cells so that healthy cells are less severely affected.Many critical properties of antibodies cannot be predicted from sequence alone and so building computational models is seen as a cost-effective aid in antibody design, maturation and formulation processes. Specifically, some important properties that can be estimated from structural models are antigen affinity (avidity), stability and aggregation propensity. Due to the high structural similarity between different antibodies and recently improved methods, structure prediction has been shown to give accurate antibody structure models.The second Antibody Modeling Assessment (AMA-II) is a community-wide experiment organized by Pfizer and Janssen R&D in 2013 to assess the state of the art in antibody structure modeling. The participants included prediction teams from Dassault Syst??mes BIOVIA (previously Accelrys), Chemical Computing Group, the Jeffrey Gray Lab at Johns Hopkins (Rosetta Antibody), Hiroki Shirai and Astellas Pharma, Inc., Macromoltek, Schr??dinger and the fully automated PIGS (Prediction of Immunoglobulin Structure) server. The goal was to predict the 3D structure of 11 antibody variable domain targets given their amino acid sequence. The targets covered different antigen binding site conformations and represented a variety of species including human, mouse and rabbit. It was a blind-prediction study. The participants did not know the actual X-ray structures during the prediction phase.Our team consisting of Marc Fasnacht, Ken Butenhof, Anne Goupil, Francisco Hernandez-Guzman, Hongwei Huang and Lisa Yan has recently published a paper entitled "Automated Antibody Structure Prediction using Accelrys Tools: Results and Best Practices" in "Proteins: Structure, Function and Bioinformatics" (Wiley Online Library). The main BIOVIA (previously Accelrys) tool used in the study was the Discovery Studio life sciences modeling and simulation application.During the prediction phase, each team member chose to build their targets using either a single, chimeric or multiple template approach for the framework region. Then the hypervariable loop regions in the initial models were rebuilt by grafting the corresponding regions from suitable templates onto the model. The templates were selected by Discovery Studio, but with flexibility of human intervention. In the post-analysis phase, the team carried out a systematic study of the modeling methods employed during the "blind phase" for framework modeling, e.g. single, chimeric and multiple template approaches using a fully automated template selection process. In addition, they also examined the factors affecting the refinement of the complementary determining regions (CDRs).The analysis of the models shows that Discovery Studio enables the construction of accurate models for the framework and the canonical CDR regions, with backbone root-mean-square deviations (RMSDs) from the X-ray structures on average below 1 ?? for most of the regions. The fully automated multiple-template approach matched or exceeded the best manual results. The antibody model assessment shows that the submitted models are high quality, with local geometry assessment scores similar to those of the target X-ray structures.Our team"s approach as published in Proteins demonstrates the reliability of automated template selection for framework and CDR regions based on a curated database. The automated methods in our study generate models as good as or better than those with manual intervention. You can access the paper here: "Automated Antibody Structure Prediction using Accelrys Tools: Results and Best Practices"