Research from the US shows that artificial intelligence (AI) can accurately predict how proteins interact with one another. 

Proteins rarely act alone. Instead, they bind to other proteins, forming protein complexes, to carry out essential biological processes like cell communication, immune responses and metabolic processes. Understanding these interactions is crucial, but experimentally determining how proteins fit together is slow, expensive and technically challenging.  

Scientists at UT Southwestern Medical Center, Texas, developed an AI-based approach that can predict whether two proteins interact and how they bind to each other. The AI model was trained with data from known protein structures and interactions from yeast, a classic organism for studying protein-protein interactions. 

By learning patterns from the data, the AI was able to predict protein interactions with high accuracy, including interactions that had not been discovered. The team identified more than 100 new probable protein complexes and provided models of how proteins interact in more than 700 complexes that were not yet well-characterised. 

This AI-based approach, published in Science, will be able to predict protein-protein interactions, potentially accelerating biomedical research and the identification of drug targets, reducing the number of studies needed for pre-clinical testing, including those that use animals.