How does AlphaFold help scientists study protein structures?

AlphaFold is an artificial intelligence program developed by DeepMind that revolutionized the study of protein structures by predicting their complex three-dimensional shapes based solely on their amino acid sequences.12

Because a protein's function is determined by its specific fold, and the number of potential configurations for even small proteins is astronomically high, predicting these structures was a long-standing challenge in molecular biology. AlphaFold2, the program's advanced iteration, utilizes a neural network architecture known as a transformer, trained on vast databases of known sequences and structures.12

By achieving an accuracy of 90 percent—comparable to experimental methods—AlphaFold has allowed scientists to map the structures of nearly 200 million known proteins across the "entire protein universe". This capability significantly accelerates research, enabling scientists to develop more effective drugs that target specific proteins associated with diseases.12