In the era of artificial intelligence, structural biology is undergoing a major shift in its methods and priorities. With technological improvements, artificial intelligence tools are able to significantly speed up the process of solving biological structures, saving time and resources. This progress is prompting many research groups to rethink their strategies, and direct their efforts towards new and innovative areas.

Certainly, AI has revolutionized structural biology as well as in other fields, and it has made the methods approaching problems differently than used to be. But someone must bear in mind that solving a biomacromolecule structure just one of step in a bid to answering important, fundamental, and unknown questions in biology. Beyond that structure determination, there are many mechanistic and structural details linked to biological functions that need to be extracted, verified, and well presented.

AlphaFold does a great job for predicting globular protein structures, many structured protein modules, and to a less extent, the protein-protein/nucleic acid complexes from their primary sequences. The success of AlphaFold hinges on substantial pool of well-determined protein structures in the database, and implementation of AI to extract a rigorous structure-primary sequence correlation that was hardly aware by human being.

Therefore, I’d like to stress that AI will not be able to do the trick if the structure is barely found or does not even exist in the database; for example, the intrinsically disordered proteins (IDPs) which have no concrete structures at all and can sample tremendously large conformational space. These disordered regions are commonly found in many proteins essential for transcriptional control, cell signalling, and many of them are highly associated with human diseases. Such conformationally heterogenous system is no longer predictable using AlphaFold; or I should state in different way, why we need to predict only one structure running up against the true nature of these conformationally heterogenous proteins. In my opinion, the direction for structural biology need to shift the gear and focus on thermodynamic behavior and structural dynamics matter in biological system.

We should feel blessing living in the era with AI but girding for. With such wonderful tool, scientists right now can ask many audacious questions that were far-fetched in the past. Again, AI is not always mighty, making use of creativity, and verifying the results from AI are extremely important, which can pave the way for science without disseminating misinformation.

Yun-Tzai