Improve Antibody Development With IgG-Trained Predictive Models
Earn a Certificate of Live Attendance
AI Webinar Explorer available after event
About this webinar
Predicting whether a therapeutic protein can be manufactured and formulated successfully typically requires a dedicated experimental campaign for each molecular format. However, the ability to apply existing antibody datasets across different molecular formats could improve developability assessments and inform antibody development decisions. This webinar explores the potential of cross-format predictive modeling using data from conventional IgG antibodies to assess VHH-Fc developability.
The session will introduce GDPa5, a 160-member VHH-Fc library profiled across ten standardized biophysical assays and examine whether models trained only on 559 IgGs can predict developability properties in a scaffold lacking both a light chain and a CH1 domain. Results demonstrate that 559 examples from a different antibody format can provide greater predictive value than 100 examples from the target format.
The webinar will also examine how prediction accuracy varies across biophysical properties. Transfer succeeds for properties encoded in local surface features, including heparin binding (Spearman ρ = 0.82), hydrophobicity (0.63) and self-association (0.62), but performs poorly for thermostability (0.16), which emerges from interdomain cooperativity absent in the target scaffold. The findings also highlight differences in model performance, with tabular neural networks better accommodating cross-format differences than linear and tree-based approaches.
Register for this webinar to explore how IgG data and predictive modeling can support VHH-Fc developability assessment and antibody development decisions.
Who should attend
- Antibody Discovery and Development Scientists
- Protein Design and Protein Engineering Scientists
- Antibody Developability and Characterization Scientists
- Computational Biology and AI/ML Scientists
- Large Molecule Discovery Scientists
- Research Managers, Directors and VPs in biopharma R&D
What you will learn
- The largest public VHH-Fc developability dataset to date and its characterization
- How IgG-trained models can predict developability properties in a different antibody format
- How model selection influences cross-format prediction performance
- Which biophysical properties are more transferable across antibody formats
Featured Speaker
Josh Moller
Josh Moller is a Senior Biological Engineer 1 specializing in AI and Machine Learning (ML) at Ginkgo Bioworks. As part of the Datapoints Antibody Developability team, they focus on the generative design of proteins and biologics, leveraging advanced computational approaches and high-throughput data generation to push the boundaries of antibody and biologic developability. Josh applies cutting-edge AI/ML models to enable predictive protein engineering and collaborates within Ginkgo’s robust R&D ecosystem to accelerate innovation in therapeutic development.
Partner for this event
Ginkgo Datapoints
Ginkgo Datapoints is an outsourced data-generation business built on Ginkgo Bioworks’ automation infrastructure as a purpose-built experimental data provider for AI-driven drug discovery. We generate experimental data for drug discovery programs at the throughput, structure, and turnaround times that modern, iterative teams actually need. To date, we’ve generated over 2.5 billion omics datapoints and more than 100,000 antibody datapoints. Built by robots, not by hand, our capabilities span antibody developability, perturbation response profiling, specialized high-throughput screening, and small molecule ADME, delivering structured, ML-ready data returns from validated, automation-driven methods.
Tangible Scientific is a tech-enabled compound management platform that removes the logistics bottleneck between design and data. From its Natick, MA facility, Tangible takes operational custody of customer compounds, handling storage, reformatting, plating, and same-day courier service to Boston-area partners including Ginkgo. AI-powered manifest reconciliation, real-time order tracking, and structured data return run through a single interface, giving discovery teams the quality of an in-house compound operation without the cost of building one. For more information, visit tangiblescientific.com.
Inductive builds virtual chemistry labs that help drug hunters design higher quality molecules, faster. Inductive’s virtual labs are designed to scale proven scientific best practices across medicinal chemistry, computational chemistry, DMPK, and safety, enabling teams to make higher-quality decisions consistently throughout discovery. Inside these virtual labs, AI chemistry assistants, predictive ADMET and PK models, and human-relevant digital organ technologies work together to help scientists evaluate more hypotheses in silico and surface key risks earlier. The most promising molecules move from the virtual lab to the wet lab in a tight feedback loop that accelerates the advancement of high-quality molecules. Inductive already powers dozens of active discovery programs, including collaborations with leading biopharma partners. For more information, please visit www.inductive.bio.