Google DeepMind Uses AI in Predicting DNA Mutations, Speeds Up Search for the Cause of Genetic Diseases

After solving the protein structure challenge, Google's AI form DeepMind has entered genetics by studying the human genome.

What is DeepMind?

DeepMind refers to an artificial intelligence technology which uses machine learning and neural networks in cracking a broad range of problems. The company DeepMind started as a London-based firm in 2010 and was acquired by Google in 2014. It is now a subsidiary of Alphabet Inc., the parent company of Google.

As an artificial neural network, DeepMind's system is organized as a network of nodes which mimics the way neurons connect to one another in the human brain. Specifically, the system uses a convolutional neural network that is organized in a similar manner as the human visual cortex, the region of the brain that processes visual information.

DeepMind was initially known for AI programs that play video games and complex board games like Go. It entered the medical field when it announced that its program called AlphaFold accurately predicted the shape of proteins, a problem which was considered as a grand challenge in biology.

Search for Disease-Causing Genes

Experts at Google DeepMind have designed an artificial intelligence program that aims to speed up research and diagnosis of rare disorders. They claim that this program can predict whether millions of genetic mutations are harmless or have the potential to cause disease.

The AI program works by checking the order of the components in human DNA strands. All living organisms are built from DNA which is composed of four blocks of chemicals called adenine (A), thymine (T), cytosine (C), and guanine (G). In a developing human embryo, the order of these letters are read in order to produce proteins that serve as the building blocks of cells and tissues of the body.

If the letters are arranged in the wrong order, the DNA is said to have undergone mutation. At this point, the cells and tissues of the body are not properly made and may lead to certain diseases. This kind of mutation is usually harmless, but it can disrupt the working of proteins and can cause diseases such as cystic fibrosis, sickle-cell anemia, cancer, and problems with brain development.

DeepMind claims that it has fine-tuned a protein model which can predict if the misspellings in human DNA are either safe to ignore or can cause disease. The new system, called AlphaMissense, can identify if the letters in the DNA will likely create the correct shape. Otherwise, it will be listed as potentially disease-causing.

With our modern technology in genetics, disease hunters only have a fairly limited knowledge of the areas of human DNA that can lead to disease. So far, they have classified 0.1% of letter changes or mutations as either benign or disease-causing. According to DeepMind computer scientist Pushmeet Kohli, the new model can increase that percentage up to 89%. It also eases the burden faced by scientists in searching for potentially disease-causing regions across billions of chemical building blocks that make up DNA.

The new tool is currently being tested by Genomics England as one of the first organizations to benefit from the new development. According to its deputy chief medical officer Ellen Thomas, this novel system helps clinical scientists in making sense of genetic data to make it useful for patients. When a type of mutation is fed on the trained AI, a score is generated reflecting how risky the genetic changes could be. However, it cannot determine how the mutations can cause any problem.