Participants are expected to have some prior exposure to computational modeling tools (e.g. Python, R, COOT, Rosetta, AutoDock Vina, etc.) but limited experience applying them to their projects. They should be comfortable using Jupyter notebooks and prepared to explore topics such as evaluating metrics, determining appropriate sampling sizes, and selecting key adjustable parameters. While this seminar does not cover ligand docking or protein-protein docking, it is well-suited for those interested in antibody modeling and, potentially, enzyme design language models.

Hands-on instructional content will be presented as Google Colab notebooks written in python. A basic understanding of general coding principles, such as typing, loops, functions, and classes, will be sufficient. It will not be required to write your own code from scratch, but a sufficient familiarity with python to understand and edit the provided notebooks will be essential to a meaningful experience.

Topics to be covered:

• Building practical experience with AI-based modeling of proteins
• A breakdown of input formats, command lines, and analysis of output
• Hands-on exercises using real-world scenarios in antibody structure prediction, developability pre-screening, immunogen solubilization, and de novo binder design
• Discussion of, and guidance on, questions like: how many models, in silico selection metrics and ranking, and how many to test in the lab
• Pipelining of protein design software and the critical use of an “oracle”​

Part 1 – Why De Novo Designs Trigger Immune Response

• Unwanted Immunogenicity fundamentals for protein engineers: What is it and why do we care
• Impact of ADAs on PK, efficacy and safety
• innate vs adaptive immune cell types and high-level roles
• T cell dependent B cell responses and ADA formation (example response to virus --> same response to a biologic) o    Not all ADAs have a negative impact or any impact
• Factors contributing to immunogenicity risk 
• Overview: drug vs patient related factors
• Sequence similarity and immunogenicity risk
• Impact of “developability” features (aggregates, polyreactivity, etc.)
• Case studies for risk assessment (interactive)
  

Part 2 – Predicting and Designing Out Immunogenicity

• In silico prediction and how they're being integrated directly into generative design loops (as scoring filters, not just post-hoc checks)
• AI/ML-driven immunogenicity prediction – training data limitations, current model accuracy, where they fail on novel modalities
• Deimmunization and epitope-deletion strategies applied to de novo scaffolds – balancing developability, stability, and immunogenicity as competing design objectives
• Using immunogenicity assessment prospectively during candidate selection and broader developability evaluation
• Case study for in silico strategy implementation in discovery (interactive)
  

Part 3 – Preclinical Validation and Practice Application

• Wet-lab assays: ex vivo T-cell activation, HLA binding, dendritic cell assays – what to run and when in the pipeline
• Tailoring assay strategies to the novel scaffolds, formats, folds, and mechanisms of generatively designed biologics
• Case studies of de novo/engineered therapeutics that hit immunogenicity issues in the clinic – lessons learned
• Regulatory expectations for immunogenicity risk assessment on novel-sequence biologics
• Combining in silico, in vitro, developability, product quality, and clinical evidence to identify unknown risks and support development decisions

By the end of the course, you should understand:

What are the consequences of unwanted immunogenicity and why it happens