AI-Driven AAV Capsid Engineering: Recent Advances in Targeted Gene Therapy Vector Design

Recent announcements demonstrate convergence around machine learning approaches for engineering adeno-associated virus capsids with improved tissue tropism and delivery efficiency. The methodologies leverage protein structure prediction integrated with computational optimization to design capsids requiring substantially lower vector doses while achieving more precise cellular targeting.

Genethon's Integrin-Targeted Muscle Capsids

Genethon's Progressive Muscular Dystrophies Team, led by Isabelle Richard, published findings in Nature Communications describing an AI-based capsid design methodology targeting Integrin Alpha V Beta 6, a receptor molecule present on human skeletal muscle cell surfaces. The workflow began with identifying this molecule of interest on muscle tissue, then modified an AAV capsid to specifically recognize this receptor using artificial intelligence tools grounded in protein structure prediction. The engineered capsids achieve 20 times dose reduction compared to natural AAVs while maintaining or improving efficacy, with simultaneous liver detargeting to reduce hepatotoxicity risk. The team applied computational predictions to assess efficiency and stability of candidate capsids before experimental validation, streamlining the transition from structural hypotheses to functional vectors.

Dyno Therapeutics' CNS and Ocular Platforms

Dyno Therapeutics has deployed machine learning algorithms for sequence-level AAV capsid design across multiple tissue tropisms. The company's bCap1 capsid demonstrates enhanced blood-brain barrier penetration for central nervous system applications while reducing liver targeting compared to natural AAV serotypes. Concurrent work optimized capsids specifically for ocular gene therapy delivery, with the company announcing critical improvements to its machine learning algorithms for capsid sequence design. Roche has committed up to 1 billion dollars in long-term collaboration with Dyno, indicating validation of the platform's clinical potential across indications.

Convergent Engineering Methodologies

Contemporary capsid engineering integrates three complementary approaches: directed evolution, rational design informed by receptor biology, and machine learning-based computational optimization. For muscle tissue applications, next-generation myotropic capsids enable uniform skeletal and cardiac transduction at substantially lower intravenous doses through liver detargeting strategies. For CNS targeting, receptor-informed engineering utilizing transferrin receptor 1 or alkaline phosphatase as ligands has produced significant gains in blood-brain barrier penetration with improved cross-species translation. Quantitative benchmarks demonstrate engineered capsids routinely deliver multi-fold improvements in potency and biodistribution relative to natural serotypes.

Broader Infrastructure Development

An NIH-funded multidisciplinary initiative has produced a toolkit comprising dozens of engineered AAV vectors capable of targeting specific cell types in the brain and spinal cord including excitatory and inhibitory neurons, cortical and striatal subtypes, spinal cord motor neurons, and brain endothelial cells. The platform underwent validation in live mammalian systems and human surgical brain specimens, demonstrating access to previously elusive cell populations in the prefrontal cortex and spinal cord motor regions. Resources are accessible through Addgene with accompanying protocols, indicating broader community adoption.