Digital Tbucket Tank (DTT)

Artificial intelligence improves viruses for gene therapy

Dependoviruses or parvoviruses "associated" with adenoviruses (AAVs) are very useful tools in the US Gene therapy. This is because they can transfer DNA into the cell and are harmless to humans. Therefore, they are used as carriers of the genetic information needed to fight diseases.

Image source: Pixabay

However, there are serious limitations that mean that their use is currently severely restricted and not all patients can use them, so not everyone can receive gene therapy. The first of these limitations is the limited ability of AAV, to attach to cells. The second limitation is the human immune system. It is estimated that 50-70% of people against one AAV infection are resistant because they have already come into contact with some form of this virus. In their case, gene therapies don't work because the immune system has the time to do that Virus to destroy before it enters the cell and with it the genetic material needed to carry out the therapy. One of the most important fields of research in gene therapy is therefore the attempt to Immune system to outsmart.


Dr. George Church of Harvard University worked with Google Research and Dyno Therapeutics one Deep learning technique used to design very different variants of the capsid (protein shell) of the AAV virus. Researchers focused on the viral genome sequences that code for a key protein segment that plays a central role in the infection of target cells and the recognition of the virus by the Immune system spielt.


The specialists showed that through the use of artificial intelligence it is possible to differentiate a large number of them Capsids that can then be tested for their ability to evade the immune system's attack. The researchers started with a small amount of data on one Capsidto target 200.000 variants.


Our research clearly shows that we are familiar with machine learning can design a huge number of variants, far more than exist in nature. We continue to refine our technology to not only create carriers that can withstand attacks by the immune system, but also to attach to selected tissue types more efficiently and selectively, "said Eric Kelsic, PhD, Director and Co-Founder of Dyno Therapeutics.
From a paper published in Nature, we learn that a preliminary evaluation of the capsids designed by the AI ​​found that almost 60% could work. This is a significant step forward. Random mutagenesis is currently used to differentiate capsids, with the percentage of usable capsids being less than 1%.
The more we deviate from the natural look of the AAV, the more likely the immune system won't recognize it, adds Sam Sinai, Ph.D., the other founder of Dyno Therapeutics, who led the team that ran the Computer modeling performed. The key to success, however, is making a capsid that can stably carry the DNA payload. Conventional methods of obtaining one Capsids are very time and resource intensive, and very few remain usable Capsids receive. Here, however, we can quickly find a great variety AAV capsids win, which is the basis for the further development of Gene therapies are available to more people. "