The many successes of AlphaFold2 (AF2) have inspired methods to predict multiple protein conformations, many of which have biological significance. These methods assume that AF2 uses coevolutionary information to predict alternative protein conformations, but they perform poorly on fold-switching proteins, which remodel their secondary structures and modulate their functions in response to cellular stimuli. Here, we present a method designed to leverage AF2’s learning of protein structure more than coevolutionary inference. This method called CF-random-outperforms other methods for predicting alternative conformations of not only fold switchers but also dozens of other proteins that undergo rigid body motions and local conformational rearrangements. CF-random captures multiple conformations more frequently and requires 3–10× less sampling than all other methods. It also enabled predictions of fold-switched assemblies unpredicted by AlphaFold3. Several lines of evidence indicate that CF-random works by sequence association, suggesting that training-set structures and sequences play an important role in which conformations can be predicted readily. This observation inspired a blind prediction mode for alternative protein conformations. We release CF-random for community use, specifying its strengths and limitations.
Lee et al. (Sun,) studied this question.