Computational discovery techniques have become integral in identifying new compounds from among many possible candidates in vast regions of chemical space. Transition metal complexes have many desirable characteristics in bonding and reactivity or electronic properties for functional materials, but strategies for their efficient design are much less well-developed in comparison to the discovery of simple organic molecules or ordered solid state materials. We develop new software tools that merge what works from therapeutic drug design and the machine learning communities with knowledge we’ve developed about transition metal chemistry to enable efficient and informed traversal of chemical space to discover new compounds. We’ve applied these ideas across materials science – from redesigning the building blocks of quantum dots to the discovery of new functionalizations for electrochemically active polymers for ion separation. Many of these advances hinge on our open-source code molSimplify, which enables automated discovery of transition metal complexes by interfacing with large chemical or user-built databases. More recently, we’ve introduce the first artificial neural network capable of predicting both spin-state and geometry of transition metal complexes to molSimplify, enabling prediction of transition metal complex properties at a fraction of the computational cost of typical DFT studies, enabling high-throughput screening.
Large-scale quantum phenomena in enzymes and the condensed phase. Knowledge of the mechanism by which enzymes accelerate reaction rates is the cornerstone of fundamental understanding in health and disease. Despite strides in understanding enzyme mechanism, guiding principles for the role of the greater protein environment in this rate enhancement remain under debate. Computational atomistic force field and first-principles simulation have emerged as crucial tools to augment experimentally-derived mechanistic understanding, as highlighted by the 2013 Nobel Prize for the development of mixed quantum-mechanical/molecular-mechanics (QM/MM) multiscale methods. First-principles, QM methods are essential to treat the charge transfer, polarization, and bond rearrangements in enzyme catalysis, albeit at much higher computational cost than MM methods. Our group has developed new quantitative methods to identify electronic contributions to enzyme properties. We’re now using these tools to reveal electronic contributions to rate enhancement, identify interactions that give rise to substrate specificity, and guide our understanding of where strong quantum mechanical interactions may reveal new structure-function relationships.
Electronic structure for transition metal chemistry and catalysis. Approximate density functional theory (DFT) is an essential tool for computational materials and catalysis discovery initiatives, owing to its balance of computational cost and accuracy, yet the accuracy of DFT is eroded proportionately to the novelty of the materials being studied (e.g., correlated electrons in transition metals). In order both for discovery efforts to become predictive and large-scale electronic structure in enzymes to provide meaningful results, these errors must be both understood and corrected. Presently available DFT functionals suffer from a mixture of errors, and standard corrections (e.g., DFT+U and range-separated hybrid functionals) improve one facet (e.g., delocalization error) while worsening another (e.g., static correlation error). Further, although energetic errors may be corrected, density errors may worsen or not improve sufficiently. We are currently developing novel correction schemes that simultaneously alleviate numerous errors by separately treating fractional spin concavity and fractional charge convexity. We also address uncertainty in DFT through an artificial neural network that can predict the sensitivity of key electronic properties to changes in functional choice.
Direct density functional theory (DFT) simulation is the critical bottleneck in high-throughput computational screening for inorganic materials and molecular transition metal complexes. These calculations are computationally costly and the calculated properties are sensitive to the exchange–correlation functional employed. Screening efforts for molecular inorganic systems are complicated by the uncertain spin-state ordering of hypothetical complexes and difficulty in obtaining good initial guess geometries.
