Application of Artificial Intelligence Approaches to Genome-targeted Therapeutic Design

Over 3% of the UK population—around 2 million people—carry one of 4,500 monogenic disorders, yet only 1 in 20 rare diseases have an approved treatment. This project builds AI models to predict how well genome-targeted therapies such as antisense oligonucleotides (ASOs) and CRISPR-based editing will work, and whether they will be toxic, before they are tested in animals or people. The core problem is that each of these 4,500 conditions has a distinct genetic defect, so developing a separate therapy for each one through conventional preclinical screening is prohibitively slow and expensive. The researchers’ insight is that many of these therapies can be repositioned across different disorders simply by changing the target sequence—a design step that machine learning could automate. By training models on large datasets of therapeutic sequences and their biological effects, the team aims to predict efficacy and toxicity in silico, dramatically cutting the need for extensive lab screening. If successful, this approach could lower development costs enough to make genome-targeted treatments accessible for the vast majority of rare genetic conditions that currently have no effective therapy. The work is applied, not fundamental science—it directly targets a bottleneck in therapeutic development.