DNA Secrets: Artificial Intelligence Decodes Human Mutations

Google’s subsidiary DeepMind has developed an artificial intelligence called “AlphaMissense”. This could predict the effects of point mutations in our DNA. Until now, it has not been clear whether these mutations are harmful, beneficial, or neutral.

AlphaMissense has now shed light on the issue, scinexx.de reports.

Point mutations arise from transcription errors during cell division or from external influences and can lead to diseases. Examples include sickle cell anemia, certain types of cancer, and a form of amyotrophic lateral sclerosis (ALS). But not all point mutations are harmful, many have no consequences.

AlphaMissense is based on three AI systems: one detects common and rare missense mutations based on existing gene databases. The second system is a linguistic model that can decode the DNA base code. The third component is a variant of AlphaFold, an artificial intelligence that can predict protein folding.

These three systems work together and allow AlphaMissense to make predictions. The AI ​​is supposed to detect whether a point mutation disrupts a protein and thus makes you sick or is harmless.

In initial tests, AlphaMissense performed as well or better than all existing models. Artificial intelligence was able to identify 71 million changes in the genetic code as missense mutations. According to expectations, 32% of them may be pathogenic, and 57% of them may be harmless.

The development of AlphaMissense is a major step in genetic research. The ability to predict the effects of point mutations allows potentially harmful mutations to be detected early. This could lead to the discovery of new disease genes and improved diagnosis of rare genetic diseases.

However, AlphaMissense predictions are just structure-based predictions. It remains to be verified whether a mutation classified as pathogenic actually causes disease.