Rizki Mardian
Principal Scientist, Protein Design and Informatics GSK
Seminars
Thursday 27th August 2026
Multi-Modal and Multi-Objective Optimization for Functional and Developable Antibody Design
11:30 am
- Computational drug discovery is a multi-layered challenge spanning many platforms and modalities; needing a scalable, integrated decision system. Value comes not just from accurate predictive oracles, but from the strategy for using them.
- The core bottleneck is data fit and quality, not volume. Both the lead optimization and developability prediction cases point to the need for data tailored to the specific projects. Models must not only predict accurately, but stay grounded in the constraints of their use case, and be able to guide new data generation that improves predictive performance over time.
- Automation infrastructure, i.e., design automation, is what turns good models into real acceleration. Without robust data/model pipelines and workflow automation linking generative models, predictive oracles, and lab feedback, even accurate individual models fail to translate into faster discovery.
