Recently, researchers from the Broad Institute of MIT and Harvard University published a research paper titled "Systematic multi-trait AAV capsid engineering for efficient gene delivery" in Nature Communications. The study developed a general machine learning method for systematically designing multi-feature AAV capsids, Fit4Function, which generates reproducible screening data by utilizing AAV capsid libraries that uniformly sample the manufacturable sequence space to train accurate sequence-to-function models, thereby helping to design AAV protein shells (capsids) with multiple ideal features, such as the ability to deliver genes to specific organs or achieve gene delivery in multiple species, thereby helping to accelerate the engineering of AAVs for gene therapy.
In this study, the research team used their machine learning approach to design capsids for a commonly used AAV serotype, AAV9, that would target the liver more effectively and could be easily manufactured. They found that about 90% of the AAV capsids predicted by the machine learning model did successfully deliver cargo to human liver cells and met five other key conditions. They also found that the machine learning model could correctly predict the behavior of macaque proteins even when trained only on mouse and human cell data. The findings suggest that this new approach could help scientists more quickly design AAVs that work across species, which is critical for bringing gene therapy to humans.
| Cat.No. | Product Name | Price |
|---|---|---|
| VLP-AAV006 | AAV9 Virus-Like Particles (Empty Capsids) | Inquiry |
| AAV00274Z | AAV9-cTNT-GFP | Inquiry |
| AAV00286Z | AAV9-CMV-RLuc | Inquiry |
| AAV00534Z | AAV9-hSyn-NULL | Inquiry |
| AAV00119Z | AAV9-CMV-Luc | Inquiry |
| AAV00271Z | AAV9-Syn-iCre | Inquiry |
| AAV00278Z | AAV9-CAG-GCaMP6m | Inquiry |
| AAV00289Z | AAV9-EF1α-mCherry | Inquiry |
| AAV00290Z | AAV9-CAG-Gluc | Inquiry |
| AAV00294Z | AAV9-Syn-RFP | Inquiry |
| AAV00405Z | scAAV9-CAG-GFP | Inquiry |
| AAV00258Z | AAV9-CAG-Cre | Inquiry |
| AAV00259Z | AAV9-CAG-Cre-GFP | Inquiry |
| AAV00285Z | AAV9-tMCK-mCherry | Inquiry |
| AAV00532Z | AAV9-CAG-Cre-2A-GFP | Inquiry |
Traditional methods for designing AAVs involve generating large libraries containing millions of capsid protein variants and then testing them in cells and animals in several rounds of screening. This process is expensive and time-consuming, and researchers can usually only identify a few AAV capsids with specific features, making it challenging to find capsids that meet multiple criteria. Some research teams have used machine learning to accelerate large-scale AAV capsid screening and analysis, but most methods optimize one function at the expense of another. The team realized that datasets based on existing large AAV libraries were not well suited for training machine learning models. It is important to understand what is needed to better train machine learning models.
The research team first performed a preliminary round of modeling using machine learning models to generate a new medium-sized library called "Fit4Function", which contains AAV capsids that are predicted to encapsulate gene cargo well. The research team then screened AAV capsids in human cells and mice for specific functions that are critical for gene therapy in each species. They then used this data to build multiple machine learning models, each of which can predict a certain function from the amino acid sequence of the AAV capsid protein. Finally, they combined these machine learning models to create "multifunctional" AAV libraries to optimize multiple properties at the same time.
Figure 1. Systematic multi-trait protein optimization paradigm. (Eid F E, et al., 2024)
The research team combined six models to design an AAV capsid library with multiple expected functions, including manufacturability and the ability to target human cells and mouse livers, with almost 90% of AAV capsids showing all the desired functions at the same time. The research team also found that the model, trained only on data from mice and human cells, could correctly predict how AAVs would be distributed to different organs in macaques, suggesting that these AAVs do this through a cross-species conversion mechanism. This could mean that in the future, gene therapy researchers could more quickly identify AAV capsids with a variety of desirable properties for humans.
The research team said the machine learning model developed in the study is expected to help other research teams create gene therapies that target the liver or specifically avoid targeting the liver. It is hoped that other research teams will be able to use this method to generate their own machine learning models and AAV capsid libraries, which together can form a machine learning map to predict the performance of AAV capsids on dozens of traits, thereby accelerating the development of gene therapies.
Reference
Eid F E, et al. Systematic multi-trait AAV capsid engineering for efficient gene delivery. Nature Communications, 2024, 15(1): 6602.