Machine learning assisted AAV capsid bioengineering for enhanced EV loading and tissue specificity

In the field of synthetic biology, scientists follow engineering paradigms to redesign biological systems with applicability in healthcare, the environment, and industry. Inspired by natural selection, directed evolution stands out as a successful approach to protein engineering, particularly when comprehensive understanding of the underlying biophysical processes is lacking. A key component of directing evolution lies in identifying suitable screens to isolate protein variants with desired attributes. However, these screens can be time-consuming and costly due to the vast sequence space to be explored. Machine learning (ML) holds the promise of reducing the sequence space by learning patterns from previously studied protein variants to predict new protein variants with enhanced capabilities. Here, we propose to develop and apply an ML-guided directed evolution strategy to engineer adeno-associated virus (AAV) capsid proteins with enhanced suitability for gene therapy. AAVs are currently the gold standard for delivering gene therapies, which hold promise for treating rare diseases. However, AAV-delivered gene therapies are limited by immunogenicity and bioavailability issues, restricting treatment to a subset of patients. Beyond its implications for AAV-based gene therapy, this work will contribute to the broader bioscience landscape by showcasing how ML can efficiently address challenges and accelerate discovery.

BBSRC priority areas
This project falls under the following UKRI-BBSRC Priorities: data-driven biology, technology development for the biosciences, new strategic approaches to industrial biotechnology and synthetic biology. We will apply methods used in data science, such as linear regression models and neural networks, to analyze the sequence-function relationship of proteins. Using transfer learning technology, we will develop a novel ML-guided directed evolution approach, which will allow us to diversify protein variants. This ML-guided directed evolution approach will further allow us to design AAV capsid proteins with enhanced extracellular vesicle (EV) loading capacity. Through optimizing the production of EV-associated AAVs (EV-AAVs), we will overcome the major bottlenecks keeping EVAAV technology from industrial scale production, cost to manufacture and achieving required yields. Finally, in the field of synthetic biology, scientists follow engineering paradigms to redesign agents found in nature, such as viruses. Following the biological engineering paradigm, we will cycle through steps of measuring, mining, modeling and manipulating during directed evolution to redesign AAVs.