Modern strategies to design metalloproteins include the development of de novo metalloproteins, directed evolution of metalloproteins, metal site modifications that significantly alter properties or function, and anchoring of known metal sites onto protein templates. Transition metal ions or cofactors embedded in a protein enable small molecule binding and catalytic properties unachievable in the absence of a metal, while allowing the tuning of the protein’s functional properties via the modification of both the primary and the secondary coordination sphere. Computational tools can aid in the development of sequences and structures favorable for metal cofactor incorporation into de novo or template-based metalloprotein design. These tools typically identify sequence patterns or side chain arrangements that strongly correlate to metal binding, but their use of sequence-specific information make them non-ideal for metalloprotein engineering applications, where the sequence and the side chains are being modified. This study has performed an analysis on ten naturally occurring metal binding sites within experimentally determined protein structures, focusing solely on the backbone atom position and orientations. The encoding of the backbone atom structure was incorporated into thirteen features for a machine-learning classification algorithm that was able to distinguish between various classes of metal binding sites as well as between sites that did not bind metal ions with an accuracy of up to 97 %. Additionally, clustering analysis has been employed to identify geometrically distinct classes of these metal binding sites. The focus on backbone structure alone make this methodology a high-throughput method to identify sites that can be mutated to bind a metal ion.

Machine learning classification of metal-binding sites in protein structures for in silico metalloprotein engineering

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.