What To Know
- Methods such as homology modeling, where the structure of a protein is predicted based on sequence similarity to proteins of known structure, and molecular dynamics simulations, which use physics principles to simulate protein folding, have been developed.
- In the field of pharmacology, ligands are often chemical compounds designed to bind to specific target proteins, such as cellular receptors, in order to modulate their function and treat diseases.
- This expanded capacity of the AlphaFold program, considered 50% more accurate than current software methods in the prediction of protein structures and their interactions, thus opening new possibilities in diverse fields such as medicine, agriculture and biotechnology.
DeepMind’s announcement of the third version of its AlphaFold software marks a major breakthrough in the field of structural biology and artificial intelligence (AI). AlphaFold3, based on machine learning, aims to accurately model how proteins fold, opening the door to a better understanding of fundamental biological processes.
The complexity of predicting troteic structures
Proteins are chains ofamino acids that must fold into a precise three-dimensional shape to perform their specific biological functions. This folding process is fundamental to life, as it determines how proteins interact with other molecules and perform essential tasks, such as enzyme catalysis, molecular transport, cell signaling, and immune response. Understanding their three-dimensional structure is thus crucial to understanding how they function at the molecular level. However, predicting it from its amino acid sequence is a major scientific challenge. The amino acid sequence, also known as the primary sequence, is indeed a long linear chain that must fold into a complex and unique configuration to become functional. The folding process is influenced by various chemical and physical forces, such as hydrophobic interactions, hydrogen bonds, and Van der Waals forces. In addition, folding occurs in a complex cellular environment where interactions with other molecules and local conditions can also play a role. Proteins can therefore adopt a wide variety of structures. Furthermore, small variations in the amino acid sequence can lead to significant differences in the final structure.
The challenges of prediction by modeling
This complexity has long made it difficult to accurately predict protein structures from primary sequence alone. Traditionally, experimental techniques such as X-ray crystallography, nuclear magnetic resonance (NMR), and cryo-electron microscopy have been used to determine protein structures. Although these methods provide detailed information, they are expensive, laborious, and often limited to specific types of proteins. Computational approaches to protein structure prediction have attempted to fill this gap. Methods such as homology modeling, where the structure of a protein is predicted based on sequence similarity to proteins of known structure, and molecular dynamics simulations, which use physics principles to simulate protein folding, have been developed. However, these methods have significant limitations in terms of accuracy and scope. The development of artificial intelligence (IA) has, however, offered new perspectives in this area.
Credits: CHRISTOPH BURGTEDT
The power of AlphaFold3
AlphaFold, developed by DeepMindis a software designed precisely to model protein folding. Until now, previous versions focused mainly on predicting protein structures. More DeepMind has however developed a new version (AlphaFold3) which also allows to predict how these proteins interact with other biological molecules such as DNA, RNA and ligands. As a reminder, ligands are molecules that bind specifically to a target protein, forming a ligand-protein complex. These molecules can be of different chemical natures, ranging from small organic molecules to metal ions to macromolecules such as polysaccharides. In the context of structural biology, ligands are often regulatory molecules or biological mediators that interact with proteins to regulate their activity. For example, in the field of pharmacology, ligands are often chemical compounds designed to bind to specific target proteins, such as cellular receptors, in order to modulate their function and treat diseases. This expanded capacity of the AlphaFold program, considered 50% more accurate than current software methods in the prediction of protein structures and their interactions, thus opening new possibilities in diverse fields such as medicine, agriculture and biotechnology. However, an important limitation is that unlike its predecessors, AlphaFold3 is not open sourcemeaning that researchers cannot publicly access its code or training data. This could limit the scientific community’s ability to customize the model or use the software for specific research projects. However, non-commercial researchers can still access AlphaFold3 through DeepMind’s AlphaFold server, which allows them to submit molecular sequences and obtain protein structure predictions in minutes. However, there is a limit of twenty jobs per day, which could limit its intensive use in certain research situations.
