Application of deep metric learning to molecular graph similarity.

Graph based methods are increasingly important in chemistry and drug discovery, with applications ranging from QSAR to molecular generation. Combining graph neural networks and deep metric learning concepts, we expose a framework for quantifying molecular graph similarity based on distance between learned embeddings separate from any endpoint. Using a minimal definition of similarity, and data from the ZINC database of public compounds, this work demonstrate the properties of the embedding and its suitability for a range of applications, among them a novel reconstruction loss method for training deep molecular auto-encoders. Finally, we compare the applications of the embedding to standard practices, with a focus on known failure points and edge cases; concluding that our approach can be used in conjunction to existing methods.

Explicit treatment of hydrogen bonds in the universal force field: Validation and application for metal-organic frameworks, hydrates, and host-guest complexes.

A straightforward means to include explicit hydrogen bonds within the Universal Force Field (UFF) is presented. Instead of treating hydrogen bonds as non-bonded interaction subjected to electrostatic and Lennard-Jones potentials, we introduce an explicit bond with a negligible bond order, thus maintaining the structural integrity of the H-bonded complexes and avoiding the necessity to assign arbitrary charges to the system. The explicit hydrogen bond changes the coordination number of the acceptor site and the approach is thus most suitable for systems with under-coordinated atoms, such as many metal-organic frameworks; however, it also shows an excellent performance for other systems involving a hydrogen-bonded framework. In particular, it is an excellent means for creating starting structures for molecular dynamics and for investigations employing more sophisticated methods. The approach is validated for the hydrogen bonded complexes in the S22 dataset and then employed for a set of metal-organic frameworks from the Computation-Ready Experimental database and several hydrogen bonded crystals including water ice and clathrates. We show that the direct inclusion of hydrogen bonds reduces the maximum error in predicted cell parameters from 66% to only 14%, and the mean unsigned error is similarly reduced from 14% to only 4%. We posit that with the inclusion of hydrogen bonding, the solvent-mediated breathing of frameworks such as MIL-53 is now accessible to rapid UFF calculations, which will further the aim of rapid computational scanning of metal-organic frameworks while providing better starting points for electronic structure calculations.

Extension of the Universal Force Field for Metal-Organic Frameworks.

We have extended the Universal Force Field for Metal-Organic Frameworks (UFF4MOF) to cover all moieties present in the most extensive framework library to date, i.e., the Computation-Ready Experimental (CoRE) database (Chem. Mater. 2014, 26, 6185). Thus, we have extended the parameters to include the fourth and fifth row transition metals, lanthanides, and an additional atom type for sulfur, while the parameters of original UFF and of UFF4MOF are not modified. Employing the new parameters significantly enlarges the number of structures that may be subjected to a UFF calculation, i.e., more than doubling accessible MOFs of the CoRE structures and thus reaching over 99% of CoRE structure coverage. In turn, 95% of optimized cell parameters are within 10% of their experimental values. We contend these parameters will be most useful for the generation and rapid prototyping of hypothetical MOF structures from SBU databases.

Multiple-component covalent organic frameworks.

Covalent organic frameworks are a class of crystalline porous polymers that integrate molecular building blocks into periodic structures and are usually synthesized using two-component [1+1] condensation systems comprised of one knot and one linker. Here we report a general strategy based on multiple-component [1+2] and [1+3] condensation systems that enable the use of one knot and two or three linker units for the synthesis of hexagonal and tetragonal multiple-component covalent organic frameworks. Unlike two-component systems, multiple-component covalent organic frameworks feature asymmetric tiling of organic units into anisotropic skeletons and unusually shaped pores. This strategy not only expands the structural complexity of skeletons and pores but also greatly enhances their structural diversity. This synthetic platform is also widely applicable to multiple-component electron donor-acceptor systems, which lead to electronic properties that are not simply linear summations of those of the conventional [1+1] counterparts.

Photoinduced Charge-Carrier Generation in Epitaxial MOF Thin Films: High Efficiency as a Result of an Indirect Electronic Band Gap?

For inorganic semiconductors crystalline order leads to a band structure which gives rise to drastic differences to the disordered material. An example is the presence of an indirect band gap. For organic semiconductors such effects are typically not considered, since the bands are normally flat, and the band-gap therefore is direct. Herein we show results from electronic structure calculations demonstrating that ordered arrays of porphyrins reveal a small dispersion of occupied and unoccupied bands leading to the formation of a small indirect band gap. We demonstrate herein that such ordered structures can be fabricated by liquid-phase epitaxy and that the corresponding crystalline organic semiconductors exhibit superior photophysical properties, including large charge-carrier mobility and an unusually large charge-carrier generation efficiency. We have fabricated a prototype organic photovoltaic device based on this novel material exhibiting a remarkable efficiency.

AuToGraFS: automatic topological generator for framework structures.

Metal-organic frameworks (MOFs) and covalent organic frameworks (COFs) are recently notable examples of highly porous polymer frameworks with a raft of potential applications. Synthesis of these compounds is modular, with "connectors" and "linkers" able to be replaced almost at will in the fabrication of isoreticular frameworks (frameworks with the same underlying topology). The range of components available to form such framework structures is vast, leading to a "combinatorial explosion" problem in predicting which framework compounds might have a set of desired properties. Computational investigations can be used in both predictive and explanatory roles in this research but rely on accurate structural models. In this work, we present our software, AuToGraFS, Automated Topological Generator for Framework Structures, and show some of its advanced functionality in "computational reticular chemistry". AuToGraFS is linked to a fully featured force field to produce fully optimized structures of arbitrary frameworks. AuToGraFS, including a graphical user interface, is publicly available for download.