This study presents a novel approach to understanding the structure–property relationships in non-stoichiometric materials and interstitial alloys using graph neural networks (GNNs). Specifically, we apply the crystal graph convolutional network (CGCNet) to predict the properties of transition-metal carbides, Mo₂C and Ti₂C, and introduce the crystal graph explainer (CGExplainer) enabling model interpretability. CGCNet outperforms traditional human-derived interatomic potential models (IAPs) in prediction accuracy and data efficiency, with significant improvements in the ability to extrapolate properties to larger supercells. Additionally, the CGExplainer tool enables detailed analysis of the relative spatial positioning of atomic ensembles, revealing key atomic arrangements that govern material properties. This work highlights the potential of GNN-based approaches for rapidly discovering complex structure–property relationships and accelerating the design of materials with customized properties, particularly for alloys with variable atomic compositions. Our methodology offers a robust framework for future materials discovery, extending the applicability of GNNs to a broader range of materials systems.

We constructed a simple atomistic potential capable of accurately reproducing the energetics of the carbon vacancy arrangements in cubic Mo₂C and Ti₂C obtained from density functional theory (DFT) calculations.

When producing stable electrodes, polymeric binders are highly functional materials that are effective in dispersing lithium-based oxides such as Li₄Ti₅O₁₂ (LTO) and carbon-based materials and establishing the conductivity of the multiphase composites. Nowadays, binders such as polyvinylidene fluoride (PVDF) are used, requiring dedicated recycling strategies due to their low biodegradability and use of toxic solvents to dissolve it. Better structuring of the carbon layers and a low amount of binder could reduce the number of inactive materials in the electrode. In this study, we use computational and experimental methods to explore the use of the poly amino acid poly-L-lysine (PLL) as a novel biodegradable binder that is placed directly between nanostructured LTO and reduced graphene oxide. Density functional theory (DFT) calculations allowed us to determine that the (111) surface is the most stable LTO surface exposed to lysine. We performed Kubo–Greenwood electrical conductivity (KGEC) calculations to determine the electrical conductivity values for the hybrid LTO–lysine–rGO system. We found that the presence of the lysine-based binder at the interface increased the conductivity of the interface by four-fold relative to LTO–rGO in a lysine monolayer configuration, while two-stack lysine molecules resulted in 0.3-fold (in the plane orientation) and 0.26-fold (out of plane orientation) increases. These outcomes suggest that monolayers of lysine would specifically favor the conductivity. Experimentally, the assembly of graphene oxide on poly-L-lysine-TiO₂ with sputter-deposited titania as a smooth and hydrophilic model substrate was investigated using a layer-by-layer (LBL) approach to realize the required composite morphology. Characterization techniques such as X-ray photoelectron spectroscopy (XPS), atomic force microscopy (AFM), Kelvin probe force microscopy (KPFM), scanning electron microscopy (SEM) were used to characterize the formed layers. Our experimental results show that thin layers of rGO were assembled on the TiO₂ using PLL. Furthermore, the PLL adsorbates decrease the work function difference between the rGO- and the non-rGO-coated surface and increased the specific discharge capacity of the LTO–rGO composite material. Further experimental studies are necessary to determine the influence of the PLL for aspects such as the solid electrolyte interface, dendrite formation, and crack formation.

The properties of periodic solid-state materials are frequently modified by the inclusion of interstitial atoms deposited pseudo-randomly throughout the crystal lattice. Accurately calculating overall lattice properties currently requires time-intensive, high-level quantum (typically DFT) calculations, often on 1000s of stochastic lattices. 'Interatomic potential models' (IPM) can mitigate such computational burdens by awarding discrete coefficients to common interstitial geometric ensembles within the wider lattice, and then summing these in a simple regression. Human-designed IPMs however typically take years to be developed. Herein, using molybdenum carbide (Mo lattice, C interstitial added atom) as a model, we show a Crystal Graph Neural Network (CGNet) workflow that can derive near-perfect IPMs from <300 DFT inputs in seconds. Our pipeline (CGNet+CGExplainer) demonstrates high potential for generality in exploring lattice structure-property relations rapidly while maintaining accuracy and interpretability.