DeepMind’s AI Makes Gigantic Leap in Solving Protein Structures
Proteins have been studied by scientists for a long time. They represent the building blocks of life and only follow the guidelines of their own genetic makeup. For example, the human body needs protein to do everything. It is necessary by being able to hear all the way to make your hair grow, and everything in between. The only way to see how a protein will function is by letting it grow and observe what it does in its final form, which is comprised of 20 amino acids in a collective third-dimensional form. Those shapes help determine whether that specific cell will be healthy or not, and its main purpose.
Scientists have wanted to find out the meaning of each protein before it develops to further research and help mankind. Currently, scientists mostly use cryo-electron microscopy (cryo-EM) technology to try and determine these protein structures. More companies have been intrigued by this work and have wanted to get into the field, Google is one of them. Google is known for its artificial intelligence (AI) work but wants to reach higher goals, programs that are comparable to human intelligence. Uncovering how to determine protein structures is just another part of their long-term plan for their AI capabilities. That being said, Google has created a department DeepMind that has uncovered a new technique in determining the shape of a protein by reading its amino acids through AI. A program AlphaFord, managed by DeepMind, illustrates what a protein should look like in its final stage by using AI to determine the diverse folds for each protein from the distance of each amino acid in the structure.
Each new program that works on protein structures is required to test their techniques to see how successful it is. After obtaining the required protein structures that have been predetermined by other scientists, they only have a number of weeks before they must return their work. AlphaFord exceeded expectations with the returning results. On a scale of 100, they had received a 90 on accuracy while other programs only averaged a score of 75. However, it has been brought to attention by researchers that the protein samples given to AlphaFord are easier to detect than others. Nonetheless, this new program is extremely close or debatably has answered the protein structure problem that has astounded scientists for so long.
DeepMind is going to release the information on how their AI program works to help humanity grow and allow more exploration for scientists. These findings can help humanity by furthering the understanding of cells, determine genes that later produce viruses, and aid the development of drugs to treat people from illnesses.
Read the Full Article at Nature: DeepMind’s AI Makes Gigantic Leap in Solving Protein Structures