The protein folding problem in the "post-AlphaFold era"

Currently, the protein structure prediction method, using deep learning artificial intelligence (AI) represented by AlphaFold, can predict protein structures at atomic resolution within seconds. To date, nearly every known protein structure on Earth has been predicted. Scientists can even design functional proteins de novo with novel topologies that do not exist in nature. AlphaFold has changed structural biology and the entire field of life sciences. Life science research seems to have entered the post-AlphaFold era. Although the development of AlphaFold has created new possibilities in life sciences, it comes with new problems and challenges. Protein folding is a physicochemical process that involves changes in a series of non-covalent interactions, which are affected by a range of physical and chemical factors. However, protein folding occurs in living cells, and our understanding of the process within the complex cellular environment remains limited. Notably, there is a very complex and precisely regulated proteostasis network in vivo, the imbalance of which results in protein misfolding and associated conditions, such as neurodegenerative diseases. Scientists in different fields have different expectations of the post-AlphaFold era. Herein, we review the early history of research on protein folding, particularly the contributions of Chinese scientists. We summarize protein folding problems in vitro and in vivo. The classic protein folding problem in vitro refers to how polypeptide chains fold into three-dimensional protein molecules. Following the thermodynamic hypothesis of protein folding proposed by Anfinsen, several phenomenological models have been constructed to describe alternative pathways through which a structure can undergo protein folding. The energy landscape theory has provided a framework for understanding the protein folding problem at a quantitative level. In vivo protein folding refers to the intracellular synthesis of nascent peptide chains from amino acid linking on ribosomes to the final maturation into functional proteins, involving protein folding, modification, transmembrane transport, assembly, secretion, and even degradation, and is considerably complex than the in vitro counterpart. The dynamic changes of protein conformation within the cellular environment are affected by various factors and regulated by interactions with other biomacromolecules. Thus, protein folding in vivo is a highly dynamic process affected by the organism's different temporal and spatial dimensions. As we enter the post-AlphaFold era, we should break the boundaries of traditional disciplines and comprehensively analyze the protein folding problem using a multidisciplinary approach. In the future, deep learning AI will not only be used to predict the static structure of proteins but is also expected to realize the prediction of protein dynamics in cells, which will facilitate better understanding of the regulatory mechanisms of proteostasis networks and promote the intelligent design of precision drugs. From the perspective of scientific development, a successful protein folding theory may provide a deep understanding of the activities, mechanisms, functions, and origins of protein-based life on Earth.