Deep Learning Metal Complex Properties with Natural Quantum Graphs

Abstract

Machine learning can make a strong contribution to accelerating the discovery of transition metal complexes (TMC). These compounds will play a key role in the development of new technologies for which there is an urgent need, including the production of green hydrogen from renewable sources. Despite the recent developments in machine learning for drug discovery and organic chemistry in general, the application of these methods to TMCs remains challenged by their higher complexity and the limited availability of large datasets. In this work, we report a representation for deep graph learning on TMCs – the natural quantum graph (NatQG), which leverages the electronic structure data available from natural bond orbital (NBO) analysis. This data was used to define both the topology and the information expressed by the NatQG graphs. At the topology level, two different NatQG flavors were developed: u-NatQG, with undirected edges, and d-NatQG, with edges directed along donor → acceptor orbital interactions. At the information level, the node and edge attribute vectors of both graphs contain NBO data, including natural charges and bond orders. The NatQG graphs were used to develop graph neural networks (GNNs) for the prediction of the quantum properties underlying the structure and reactivity of TMCs (e.g. HOMO-LUMO gap and polarizability). These models surpassed baselines based on traditional descriptors and performed at a level similar to, or higher than, state-of-the-art GNNs based on radial cutoffs. The results showed that the electronic structure information encoded by the models has a stronger impact on its accuracy than the geometric information. With the aim of benchmarking the GNNs, we also developed the transition metal quantum mechanics graph dataset (tmQMg), which provides the geometries, properties, and NatQG graphs of 60k TMCs.

Deep Graph Learning for Molecules and Materials

“…Kneiding and coworkers [6] presented a novel representation for TMCs, the natural quantum graph, utilizing Natural Bond Orbital (NBO) data to derive physically meaningful molecular graphs and featurize nodes as well as edges with additional information. NBO analysis yields a localized picture of the electronic structure by classifying orbitals as either lone pairs (LPs) centered on one atom, bonding orbitals (BDs) centered between two atoms and three-center bonds (3Cs) centered over three atoms.…”

“…In chemistry specifically, ML has revolutionized, among others, the fields of drug discovery [1], materials science [2], and catalysis [3] and has successfully been used to predict various properties of molecules and materials. [4,5,6] These so-called surrogate models are trained on data obtained from experiments or computationally expensive reference methods, that give access to a multitude of molecular properties. Because the references are usually electronic structure methods such as density functional theory (DFT) the field is often referred to as quantum chemistry machine learning (QCML) 1 .…”

“…[12] The leveraging of these highly expressive representations with graph neural network (GNN) models has been shown to reach state-of-the-art accuracy in the prediction of various quantum chemical properties of molecules and materials. [4,5,6] GNNs operate on the premise that global properties of molecules and materials are governed by local interactions of atoms and exploit their locality akin to how convolutional neural networks exploit information about neighboring pixels in the field of computer vision. Internally, GNNs learn a set of descriptors that is specific to the used data and prediction task, which gives them more flexibility compared to the traditional static descriptors.…”

Hybrid DFT Geometries and Properties for 17k Lanthanoid Complexes─The LnQM Data Set

“…36 The tmQM was recently extended by the tmQMg and the tmQMg-L data set. 37,38 Although the tmQM data set considerably extends the spectrum of available sets beyond organic chemistry to transition metals, there is currently no extensive data set for ML applications in the field of lanthanoids.…”

Applying Large Graph Neural Networks to Predict Transition Metal Complex Energies Using the tmQM_wB97MV Data Set

“…3 Despite the fact that homogeneous catalysts offer tremendous opportunities for selectivity and tunability that heterogeneous catalysts do not, 12,13 the field has been lacking similar large-scale, diverse data sets for homogeneous catalysis until the tmQM data set was published in 2020. 11 Additionally, to the best of our knowledge, no ML potential specifically trained on tmQM has been unveiled; Balcells et al have since released a modified version of the data set, tmQMg, 14 which was used to train various graph models. We took several GNNs that have shown promise on OC20 and OC22 and trained them on tmQM.…”