Since its launch in 2020, more than 2 million researchers have utilized the Alphafold 2 model from Google DeepMind to predict protein structures for various scientific tasks, including vaccine development and cancer treatment. This groundbreaking achievement solved a problem that scientists had been grappling with for over 50 years. Instead of resting on their laurels, the team immediately set to work on developing Alphafold 3.
The new model, unveiled in May, goes beyond predicting only protein structures. It now has the capability to predict interactions among all types of life molecules, such as DNA, RNA, and Ligands – small molecules that bind to proteins.
While the previous model, Alphafold 2, aided researchers in making significant discoveries, advancements in biology and chemistry demanded more. Alphafold 3 now encompasses all biomolecules, catering especially to drug development, as nearly half of all medicines consist of ligands. Researchers are now using Alphafold 3 to analyze the binding of new small molecules to new targets, examine protein interactions with DNA and RNA, and assess the impact of chemical modifications on protein structures.
Moreover, Alphafold 3 provides access to the Alphafold Server online service, enabling scientists to input their sequences and generate molecular complexes. Since its launch in May, this free tool has already been used to create over 1 million structures.
The new model has expanded the dataset for learning, incorporating DNA, RNA, small molecules, and other biomolecules, resulting in improved prediction accuracy. The model’s architecture has also been revamped, now utilizing a generative model based on diffusion that simplifies the processing of diverse molecules.
However, a new challenge surfaced: the model lacked training data on “disordered regions” of proteins, which led to the creation of incorrect “ordered” structures. To tackle this issue, Alphafold 3 incorporated predictions from Alphafold 2 to accurately forecast disordered interactions.
The team is optimistic that Alphafold 3 will pave the way for new breakthroughs in genomic research and drug development. With each advancement, the model continues to push boundaries, making the seemingly impossible achievable and aiding in solving increasingly complex scientific puzzles.