AI model "Eden" to generate new gene therapies and drugs

With the AI model "Eden", scientists hope to create new gene therapies and drugs that can cure the worst diseases, reports the Financial Times. The AI model is intended to derive therapies from evolution and thus be able to recognize which enzymes are suitable for gene therapy. However, observers still see many open questions that need to be answered.

Even the name "Eden" is a statement, alluding to the biblical Garden of Eden: the international team uses a huge dataset for its AI model. This includes over ten billion genes from more than 1.03 million species, most of which are from microbes. The British company Basecamp Research contributed the data. The training material had not previously been available in public databases. Nvidia is acting as an investor and partner through its venture capital arm, NVentures, as is Microsoft.

Therapy to supplement defective genes

But it is not only the scope of the data material that has apparently encouraged the researchers to allude to the Christian doctrine of creation in the name of the AI model. The approach of how people are to be helped with the findings also sounds ambitious. With AI-designed enzymes, precise, large gene insertions are to be made in humans.

In addition to gene addition, there are also methods of rewriting DNA (gene editing) and switching off problematic genes (gene silencing). The AI-generated gene can be transported into the cell using modified, harmless viruses. Alternatively, lipid nanoparticles, as with mRNA vaccines, or direct injections can be used.

These gene supplements have already been made at over 10,000 disease-relevant sites in the human genome. They supplement defective genes. In laboratory tests, cancer cells have already been killed in this way. Compared to previous methods, this insertion was achieved without DNA damage and for larger DNA segments. This prevents, among other things, changes at incorrect locations (off-target effects).

Agents against multidrug-resistant bacteria?

However, the research is also intended to help, for example, to generate peptides against multidrug-resistant bacteria. With the AI model as a unified basis, it could also become possible to develop medicine for rare diseases faster and cheaper than before.

Researchers like Fyodor Urnov from UC Berkeley point out that a new tool alone is not enough. Whether the technology can actually be applied clinically also depends on factors such as regulation, manufacturing, and insurance. Omar Abudayyeh from Harvard Medical School also expressed skepticism: the safety and efficacy of the method still need to be proven.

(mki)