"DeepMind Unveils AlphaMissense: Revolutionizing DNA Mutation Effect Prediction with AI"

 At Google, researchers at the DeepMind laboratory are deeply intrigued by predictive analysis of human DNA. They have recently developed an artificial intelligence model capable of determining whether a so-called missense mutation in the genetic code can trigger a medical condition. This project, named AlphaMissense, is an adaptation of AlphaFold, another research project by DeepMind focused on protein structure, which was in competition with Meta's ESMFold before the latter was dismantled.

As reported by the British newspaper The Guardian, the researchers published their work on September 19th in the scientific journal Science. They aim to make their AI model available on a large scale. To this end, DeepMind has open-sourced their innovation, which is now accessible on the GitHub platform. They have also released a free catalog of their millions of predictions on the Zenodo website.

AlphaMissense's analysis is particularly detailed and aligns with the goals of ESMFold. The AI model can determine whether single-point genetic mutations, where only one nucleotide has been altered during DNA replication, are likely to cause a disease or not. For example, sickle cell anemia is caused by such missense mutations.

According to DeepMind's calculations, out of the 71 million observed mutations, 57% of them are considered relatively benign, while 32% could potentially lead to a medical condition. There is still uncertainty regarding the remaining 11%, as explained by Google's laboratory in a blog post. On average, individuals carry more than 9,000 missense mutations in their DNA, but many of these mutations are still unknown to scientists.

DeepMind emphasizes that while their predictions are not intended for direct clinical use and should be interpreted alongside other sources of data, this work could enhance the diagnosis of rare genetic diseases and contribute to the discovery of new genes responsible for diseases. Scientists will need to utilize this new tool and compare Google's predictions with other analysis methods.

To date, only 0.1% of the 71 million missense mutations identified by AlphaMissense have been verified by human experts. Ben Lehner, a human genetics specialist interviewed by The Guardian, currently finds that DeepMind's AI is quite adept at identifying mutations that could lead to medical conditions. In pursuit of external evaluation, Google's researchers have collaborated with Genomics England, a British company established by the country's health ministry.