Key points are not available for this paper at this time.
Abstract Current AI tools for designing macromolecules have been struggling with DNA and RNA structure prediction and design as significantly less experimental training data are available for nucleic acids than for proteins. Therefore, developing alternative approaches remains of interest. MUMBO, a program for designing protein–protein interactions and protein–ligand-binding pockets, uses side-chain-packing algorithms to select from alternative amino acids and conformers generated on a fixed backbone with the help of rotamer libraries. MUMBO identifies the most favourable combinations based on the lowest overall energy. In order to extend the program’s capabilities to designing nucleic acids, we developed NuConf, a discrete pseudorotational angle-dependent nucleoside-specific rotamer library. We derived NuConf by statistically analysing pseudorotational and dihedral angles of more than 175,000 nucleotides from experimental structures and validated it by rebuilding more than 20,000 nucleotides in a custom dataset. Strikingly, our approach predicts nucleotides at least as accurately as amino acids. We show that the implementation of the NuConf library in MUMBO enables modelling and designing DNA and RNA sequences on a fixed backbone together with protein-nucleic acid interaction interfaces. Because its approach and format are program agnostic, NuConf can be used by other molecular design frameworks as well.
Makarova et al. (Tue,) studied this question.