Discovery and language model-guided design of hyperactive transposase

Abstract The PiggyBac transposase gene writing system has been efficiently used across biotechnological applications, however its diversity and biochemical potential remain largely unexplored. By developing a eukaryotic transposon mining pipeline, we expand the known diversity by two orders of magnitude and experimentally validate a subset of highly divergent PiggyBacs. We then fine tune a protein language model to further expand PiggyBac sequence space and discover transposons with improved activity, compatible with T-cell engineering and Cas9-guided programmable transposition. Our work illustrates how combining bioprospecting and AI-driven sequence exploration can accelerate the discovery of novel eukaryotic gene writing tools.