AI-designed proteins surpass nature in genome editing
The discovery could accelerate gene and cell therapies for cancer and rare diseases.
Researchers in Barcelona have developed synthetic proteins using generative AI that outperform natural ones at editing the human genome. The breakthrough, published in Nature Biotechnology, could transform treatments for cancer and rare genetic diseases.
The team from Integra Therapeutics, UPF and the CRG screened over 31,000 eukaryotic genomes, identifying more than 13,000 previously unknown PiggyBac transposase sequences. Experimental tests revealed ten active variants, two matching or exceeding current lab-optimised versions.
In the next phase, scientists trained a protein large language model on the newly discovered sequences to create entirely new proteins with improved genome-editing precision. The AI-generated enzymes worked efficiently in human T cells and proved compatible with Integra’s FiCAT gene-editing platform.
The Spanish researchers say the approach shows AI can expand biology’s own toolkit. By understanding the molecular ‘grammar’ of proteins, the model produced novel sequences that remain structurally and functionally sound.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!