AlphaFold 3 modeling of DNA nanomotifs: is it reliable?

Abstract

Being able to accurately predict structures is highly desirable for nanoengineering with DNA and other biomolecules. The newly launched AlphaFold 3 (AF3) provides a potential platform for this purpose. In this work, we have used AF3 to model a list of commonly used DNA nanomotifs and compared the AF3 structures with the experimentally observed structures reported in the literature. For asymmetric motifs, AF3 structures are consistent with the experimental observations; but for symmetric motifs, AF3 structures are often substantially different from experimental observations. However, the fails can be rescued if the symmetric motifs are converted into corresponding asymmetric motifs by breaking DNA sequence symmetry while maintaining the backbone symmetry. This study suggests that while AF3 is immensely helpful, we as experimentalists should use it (as it currently stands) with caution. In addition, AF3 needs further development to incorporate the existing experimental data in the training dataset for AF3. At the current stage, a hybrid approach might be beneficial: theoretical modeling softwares calculate the detailed, 3D DNA structures based on secondary DNA structures inspired by experimental observations.