The understanding of how genes and their products function inside the body—as well as how and why they occasionally go wrong—is being progressively increased by Google’s AI research division DeepMind.
The AI research group DeepMind revealed in 2021 that AlphaFold, their first digital biological neural network, is being developed. The 3D structure of proteins, which dictates the roles that these molecules play, could be predicted with accuracy by the model. Pushmeet Kohli, vice president of research at DeepMind, states, “We’re just floating bags of water moving around.” The components of life, proteins, are what distinguish us from other species. The magic of life lies in their interactions with one another.
The magazine Science named AlphaFold the innovation of the year for 2021. It was the AI research paper with the most citations in 2022. According to Kohli, “People have been working on [protein structures] for many decades and have not been able to make that much progress.” “AI then emerged.” Additionally, DeepMind made the AlphaFold Protein Structure Database publicly accessible to researchers across the globe. This database included the protein structures of nearly all organisms whose genomes had been sequenced.
For studies ranging from the creation of plastic-eating enzymes to the creation of more potent malaria vaccinations, more than 1.7 million researchers in 190 countries have used it. Understanding cancer, COVID-19, and neurological illnesses like Parkinson’s and Alzheimer’s accounted for 25% of AlphaFold’s research. With the release of AlphaFold’s next generation last year, DeepMind expanded the application of its structure prediction algorithm to biomolecules, including ligands and nucleic acids.
According to Kohli, “It has democratized scientific research.” Researchers studying a neglected tropical disease in a developing nation lacked the resources to compute a protein’s structure. They can now access the AlphaFold database with a single click and obtain these forecasts for free. For example, the Drugs for Neglected Diseases Initiative, one of DeepMind’s early collaborators, used AlphaFold to create drugs for diseases like leishmaniasis, sleeping sickness, and Chagas disease that affect millions of people but receive very little research.
The most recent innovation from DeepMind is known as AlphaMissense. The model classifies what are known as missense mutations, which are genetic changes that might cause different amino acids to be generated at certain locations in proteins. These mutations have the potential to change how the protein functions and AlphaMissense assigns a likelihood value based on whether the mutation is pathogenic or benign. According to Kohli, “It is essential for the discovery of rare genetic diseases to understand and predict those effects.” Since its debut last year, the program has successfully identified approximately 89% of all potential human misinterpretations. Previously, researchers had only clinically classified 0.1 percent of all potential variations.
According to Kohli, “This is just the beginning.” In the end, he thinks AI might result in the development of a virtual cell, which might drastically speed up biomedical research and allow biology to be studied in silico as opposed to in actual labs. “We finally have the means to understand this incredibly complex system that we call life thanks to artificial intelligence and machine learning.”
