Figure 1. STREUSEL ("This work") atomic radii are comparable with those previously reported in literature. Specifically, we highlight that our radii do not represent the extrema in the accepted atomic radii ranges.

in review, ​2020.

How Big Are Small Molecules?

Chemistry is rife with size-dependent physicochemical properties including sterics-driven chemical interactions, non-covalent interactions, enzyme docking mechanisms, and electrochemical properties. Quantifying atomic and molecular size, however, is convoluted by the definition of surface, since calculated sizes depend on model assumptions and chemical environment. For example, conventional size metrics predicated on van der Waals radii are not applicable to ion size quantification for electrolytes because the spherical approximation is not valid in highly polarized systems. Indeed, alternative size quantification methodologies, both experimental and theoretical, are necessary; several alternative approaches have been presented. Yet, independent of exact methodology, all size calculation metrics work to identify the extent to which a chemical system permeates in space. The advent of inexpensive computational developments and hardware advancements has enabled DFT-derived size calculation methods, particularly useful for emerging exotic molecules.
We present STREUSEL (Structure Topology REcovery Using Sampling of the ELectric field), a novel theoretical atomic size quantification methodology dependent on the electric field, which we have previously applied to assess the role of sterics in nucleolar stress of platinum(II) compounds (J. Am. Chem. Soc., 2019, 141, 18411).  This method is robust even for polarized systems and enables us to not only quantify atomic and molecular size, but obtain analytical interaction energetics, due to their dependence on the electric field. Indeed, we show the atomic radii recovered by STREUSEL are comparable with accepted atomic radii presented by Alvarez, Boyd, and Hoffmann, Figure 1.

The pursuit of Quantitative Structure-Flavor Relationships

The partnership of computational chemistry and computer science may be used to efficiently identify the subtle trends in diverse chemical spaces. Here, we use this marriage to pursue a quantitative relationship between molecular structure of compounds commonly found in coffee and their perceived flavors. This work ultimately has the potential to offer consumers a customizable cup of coffee, which is unobtainable in the current 7 billion USD industry.
​By way of electronic structure theory, we are able to identify atomic contributions to bulk properties, which has powerful implications in the chemistry and physics of sustainable coffee production and consumption. For example, espresso is one of the more energy expensive brewing procedures, however, its unique flavor and aromatic profile are only derived within high pressure brewing devices. Computational chemistry offers a low-cost avenue to explore alternative chemical approaches to obtaining a similar flavor profile using less energy; perhaps by altering the ionic composition of the water, which is known to impact consumer experience, or developing new filtration materials guided by the chemistry of molecular sieves, or even developing novel brewing procedures/hardware (an avenue that I am actively working on and well-placed to tackle in light of my mechanical engineering background). Even more powerful, the pairing of computational chemistry with the predictive and generative power of machine learning offers an exciting new direction for accelerated navigation of the diverse chemical space comprised by the complex molecular construction of coffee.

Figure 3. The experimental pore volumes were obtained for a subset of the CoRE MOF database. We demonstrate good agreement between our calculated pore volume and experimental values. Discrepancies between experiment and theory may be attributed to missing linker defects and occluded pores with excess solvent from synthesis.

in review, ​2020.

Novel porosity metric for metal-organic frameworks (MOFs)

​
Metal-organic frameworks (MOFs) are an emerging material in the energy storage device arena due to their nanoporous, crystalline structure, and tunability of the two basic building blocks: metal centers and organic linkages. The structural modularity of these materials renders them ideal for high-throughput materials design approaches that take advantage of the predictive power of machine learning (ML). Accuracy of ML models depends on the reliability of the labeled training data set. Therefore, improved data labels will have significant impact on successful design and property prediction of MOFs. The chemical and physical properties of a MOF are derived from the pore, which is best described via the pore volume (PV) and surface area (SA). Current methods of calculating and measuring PV and SA rely on inert gas probe molecules, yielding only the accessible PV and SA, in addition to an inherent dependence on probe molecule identity. This is detrimental to ML models, which rely on accurate predictive features for training. We have increased the accuracy and efficiency of PV and SA calculations by taking advantage of density functional theory, which yields the electrostatic potential at a discrete number of volumetric pixels within the unit cell. Using a systematic sampling approach coupled with blob and Canny edge detection algorithms we successfully demonstrate the utility of this methodology for a data set of MOFs labeled with experimental pore volumes, Figure 3. The presented methodology is independent of probe molecule identity, thus universalizing PV and SA calculations for all porous materials and removing bias from the model. 

​