Structural Biology in the AlphaFold Era: How Far Is Artificial Intelligence from Deciphering the Protein Folding Code?

Proteins are biomolecules characterized by uncommon chemical and physicochemical complexities coupled with extreme responsiveness to even minor chemical modifications or environmental variations. Since the shape that proteins assume is fundamental for their function, understanding the chemical and structural bases that drive their three-dimensional structures represents the central problem for an atomic-level interpretation of biology. Not surprisingly, this question has progressively become the Holy Grail of structural biology (the folding problem). From this perspective, we initially describe and discuss the different formulations of the folding problem. In the present manuscript, the folding problem is framed from a historical perspective, effectively highlighting the progress made in the last lustrum. We chronologically summarize the major contributions that traditional methodologies provide in approaching this multifaceted problem. We then describe the recent advent and evolution of predictive approaches based on machine learning techniques that are revolutionizing the field by pointing out the potentialities and limitations of this approach. In the final part of the perspective, we illustrate the contribution that computational approaches will make in current structural biology to overcome the limitations of the reductionist approach of studying individual molecules to afford the atomic-level characterization of entire cellular compartments.