Dominykas Bukelskis

TITLE: Building the Data Behind AI-Driven Protein Engineering Protein research is increasingly moving from discovering useful proteins in nature toward engineering proteins with properties tailored to specific applications. Machine-learning approaches can accelerate this transition by helping researchers navigate vast protein sequence spaces and prioritize promising designs. However, these approaches depend on experimental sequence–function data, which can require substantial time, effort and resources to generate. For many laboratories, building and testing hundreds or thousands of protein variants may therefore make AI-driven protein engineering appear difficult to access. Traditional workflows based on cloning, transformation and cell-based expression are powerful and well established, but can become a practical bottleneck as experimental scale increases. New experimental approaches can help lower this barrier by simplifying the path from sequence design to functional testing and making parallel evaluation of protein variants more practical. This allows to create datasets at a scale that can support computationally guided protein engineering without requiring exhaustive exploration of sequence space. By combining accessible experimental workflows with computational approaches, protein engineering can move toward a more iterative design–build–test–learn cycle—bringing data-driven protein engineering within reach of more research laboratories.