The Rise of AI in Structural Biology

Abstract

In this review, I will explore how recent protein structure prediction methods that make use of deep learning will impact the field of structural biology. I will particularly search for purposes where these methods can be applied and what are their current pitfalls. After an intensive literature study, it became clear that accurate prediction algorithms are useful to solve experimental structures faster and help divide proteins into functional domains. Structural predictions improve interpreting the molecular processes of (unknown) proteins and therefore provide insights into diseases and how to design potential treatments. That way, researchers can rapidly work out the structure of every protein in new and dangerous pathogens, speeding up the process of screening for drug targets. In addition to that, the open-source nature of these prediction methods enables scientists to continue the advances to create even more powerful software. Besides great advances, limitations of these deep learning structure prediction methods include that the predicted models only provide one conformational state of the protein and that no ligands are included in the model which is crucial data for inferring exact biological function and designing new drugs. Additionally, no folding information is present for the predicted structure. We should not state that these algorithms solved the folding problem, as they solved the prediction problem. Although scientists should remain critical of computational predictions, these technologies are still groundbreaking and will lead to more advances in the field of structural biology.