DeepMind has created an AI algorithm called AlphaMissense to predict diseases from genetic mutations

Scientists from DeepMind have developed AlphaMissense artificial intelligence that analyses genetic mutations to detect and diagnose rare diseases.

Here's What We Know

The programme makes predictions about missense mutations - one-letter errors in the genetic code. They are usually harmless, but sometimes disrupt proteins and lead to disease.

The researchers used AlphaMissense to analyse 71 million potential human single-letter mutations. The AI predicted 90 per cent accurately that 57 per cent of them were harmless and 32 per cent were harmful.

AlphaMissense predicted the pathogenicity of all 71 million possible missense mutations

Based on the results, DeepMind has created an open online catalogue of predictions to help geneticists and doctors.

According to the developers, AlphaMissense performs better than existing programmes and can identify unexplored mutation-disease relationships.

Here's How It Works

AlphaMissense is an adaptation of the AlphaFold model that predicts protein structure.

The AI analysed human DNA data to identify common and rare mutations. It also studied millions of protein sequences to understand what a "normal" protein looks like.

Based on this knowledge, the AI then estimates the risk of a particular mutation. According to the developers, this makes the analysis much simpler and faster.

Source: DeepMind