As part of the broader AI BioDesign (AIBD) initiative, the Protein Binder Flywheel integrates experimental biology and machine learning to advance biological design.

Proteins interact with other molecules through highly specific molecular recognition mechanisms, and even a single amino acid substitution can dramatically alter their binding properties. Predicting these effects remains a major challenge in protein engineering.

Our goal is to develop state-of-the-art machine learning models that predict how amino acid substitutions affect molecular binding, enabling the design of protein binders with improved affinity and specificity. The ability to design protein binders with tailored molecular recognition properties could accelerate therapeutic discovery and enable the development of next-generation biosensors. By predicting how sequence changes alter molecular interactions, our models could also help interpret genetic variants of uncertain significance, advancing our understanding of disease mechanisms and informing precision medicine.

To achieve this, we combine high-throughput experimentation with machine learning in an iterative Design–Build–Test–Learn (DBTL) cycle. Each round of experiments generates high-quality data to improve model predictions, while the models guide the design of subsequent experiments. This continuous feedback loop aims to improve predictive accuracy and generalizability across diverse protein families and molecular interactions.

‍