Generative AI Creates Superior Gene Editing Tools

BARCELONA, SPAIN – Researchers from Integra Therapeutics, in collaboration with Pompeu Fabra University (UPF) and the Centre for Genomic Regulation (CRG), have used generative AI to create synthetic proteins that significantly outperform their naturally occurring counterparts used for human genome editing.

The work, published today in Nature Biotechnology, focused on enhancing PiggyBac transposases, mobile genetic elements widely used for gene delivery due to their ability to carry large DNA payloads. This breakthrough has the potential to dramatically improve current gene editing tools for applications like CAR T-cell therapies and gene therapies.

Generative AI Expands the Genetic Toolkit

The research began with extensive computational bioprospecting across more than 31,000 eukaryotic genomes, uncovering over 13,000 previously unidentified PiggyBac sequences.

By fine-tuning a protein large language model (pLLM) called ProGen2 with these newly discovered sequences, the team was able to generate entirely synthetic proteins.

“For the first time, we have used generative AI to create synthetic parts and expand nature,” said Dr. Marc Güell, an ICREA research professor at UPF and scientific director at Integra Therapeutics. “Like the cognitive power of ChatGPT can be used to write a poem, we have used the protein-based large language models to generate new elements that comply with the physical and chemical principles of genes.”

Synthetic Enzymes Show Superior Performance

The team experimentally tested 22 synthetic variants, finding that seven outperformed the widely used natural enzyme, HyPB (hyperactive PiggyBac), in excision activity.

Crucially, one AI-designed sequence, dubbed “Mega-PiggyBac,” showed significantly improved performance in both the excision and targeted integration of DNA.

Additionally, the synthetic transposases proved highly compatible with the FiCAT gene editing platform (which fuses a Cas9 enzyme to a PiggyBac transposase for precise targeting). One synthetic sequence doubled the integration efficiency of FiCAT, highlighting its potential for use in precise genome engineering applications.

This study marks the first successful application of large language models to design synthetic PiggyBac systems, offering a “valuable framework” for developing other gene modification tools for efficient and precise genome manipulation in both biotechnology and therapeutics. The research teams plan to move forward by evaluating the AI-optimized transposases in animal models, with a key focus on ensuring high specificity to avoid off-target effects in clinical settings.