The predictive tool AlphaFold2 developed by DeepMind (Google) in 2020 using deep learning models, revealed an unequalled accuracy and is revolutionizing the field of structural biology. It consists in predicting the 3D conformation of all atoms of a protein from its linear amino acid sequence. Its performance opens up numerous perspectives in the field of molecular biology and molecular engineering applied to materials or health. The history of the different approaches that were developed to tackle this longstanding question will be revisited and the implications of this new tool, which has been available to the scientific community for the past 18 months, will be discussed as well as its current limits and perspectives. Applications to the study of chromatin assembly mechanisms will be presented.