Case Study - AlphaFold - The Power of AI and Algorithms in Predicting Protein Structures

AlphaFold is an AI program developed by DeepMind that predicts protein structures. To understand its significance, we first need to know what proteins are and why their structure is important.

Proteins are made up of chains of amino acids that fold into 3D structures. These structures determine the protein’s biological function. The “protein-folding problem” is a major challenge in biology because it’s difficult to figure out how the amino acid sequence of a protein determines its 3D structure.

Scientists have used expensive and time-consuming methods like X-ray crystallography, cryo-electron microscopy, and nuclear magnetic resonance to determine protein structures. However, these methods have identified only the structures of around 170,000 proteins, while there are more than 200 million known proteins across all life forms. If we could predict protein structures only from their amino acid sequences, it would greatly advance scientific research. That’s where AlphaFold comes in.

AlphaFold uses AI and deep learning to predict protein structures. It has competed in the Critical Assessment of Structure Prediction (CASP) competition, which challenges scientists to produce their best protein structure predictions. In 2018, AlphaFold 1 ranked first in the competition, particularly excelling at predicting structures for difficult targets where no existing template structures were available.

In 2020, AlphaFold 2 repeated this success, achieving a level of accuracy much higher than any other group. It scored above 90 out of 100 for about two-thirds of the proteins in the CASP global distance test (GDT), which measures how similar a predicted structure is to a structure determined by lab experiments, with 100 being a perfect match.

AlphaFold 2’s results were considered “astounding” and “transformational” by researchers. While it still has room for improvement, the achievement is impressive.

DeepMind trained AlphaFold on more than 170,000 proteins from a public repository of protein sequences and structures. The program uses a form of attention network, which is a deep-learning technique that helps the AI identify parts of a larger problem, then piece it together to obtain the overall solution. This training and the subsequent predictions showcase the importance of AI and Algorithms in tackling complex problems like protein folding.

AlphaFold is a prime example of how AI and Algorithms can be used to make significant advancements in scientific research. By predicting protein structures from amino acid sequences, AlphaFold opens up new possibilities for understanding biological processes and developing treatments for various diseases.