Transfer learning using attentions across atomic systems with graph neural networks (TAAG)

From bulk effective mass to 2D carrier mobility accurate prediction via adversarial transfer learning

Data scarcity is one of the critical bottlenecks to utilizing machine learning in material discovery. Transfer learning can use existing big data to assist property prediction on small data sets, but the premise is that there must be a strong correlation between large and small data sets. To extend its applicability in scenarios with different properties and materials, here we develop a hybrid framework combining adversarial transfer learning and expert knowledge, which enables the direct prediction of carrier mobility of two-dimensional (2D) materials using the knowledge learned from bulk effective mass. Specifically, adversarial training ensures that only common knowledge between bulk and 2D materials is extracted while expert knowledge is incorporated to further improve the prediction accuracy and generalizability. Successfully, 2D carrier mobilities are predicted with the accuracy over 90% from only crystal structure, and 21 2D semiconductors with carrier mobilities far exceeding silicon and suitable bandgap are successfully screened out. This work enables transfer learning in simultaneous cross-property and cross-material scenarios, providing an effective tool to predict intricate material properties with limited data.

Physics-Guided Dual Self-Supervised Learning for Structure-Based Material Property Prediction

Deep learning models have been widely used for high-performance material property prediction. However, training such models usually requires a large amount of labeled data, which are usually unavailable. Self-supervised learning (SSL) methods have been proposed to address this data scarcity issue. Herein, we present DSSL, a physics-guided dual SSL framework, for graph neural network-based material property prediction, which combines node masking-based generative SSL with atomic coordinate perturbation-based contrastive SSL strategies to capture local and global information about input crystals. Moreover, we achieve physics-guided pretraining by using the macroproperty (e.g., elasticity)-related microproperty prediction of atomic stiffness as an additional pretext task. We pretrain our DSSL model on the Materials Project database and fine-tune it with 10 material property data sets. The experimental results demonstrate that teaching neural networks some physics using the SSL strategy can afford ≤26.89% performance improvement compared to that of the baseline models.

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

Machine learning (ML) methods have shown promise for discovering novel catalysts but are often restricted to specific chemical domains. Generalizable ML models require large and diverse training data sets, which exist for heterogeneous catalysis but not for homogeneous catalysis. The tmQM data set, which contains properties of 86,665 transition metal complexes calculated at the TPSSh/def2-SVP level of density functional theory (DFT), provided a promising training data set for homogeneous catalyst systems. However, we find that ML models trained on tmQM consistently underpredict the energies of a chemically distinct subset of the data. To address this, we present the tmQM_wB97MV data set, which filters out several structures in tmQM found to be missing hydrogens and recomputes the energies of all other structures at the ωB97M-V/def2-SVPD level of DFT. ML models trained on tmQM_wB97MV show no pattern of consistently incorrect predictions and much lower errors than those trained on tmQM. The ML models tested on tmQM_wB97MV were, from best to worst, GemNet-T > PaiNN ≈ SpinConv > SchNet. Performance consistently improves when using only neutral structures instead of the entire data set. However, while models saturate with only neutral structures, more data continue to improve the models when including charged species, indicating the importance of accurately capturing a range of oxidation states in future data generation and model development. Furthermore, a fine-tuning approach in which weights were initialized from models trained on OC20 led to drastic improvements in model performance, indicating transferability between ML strategies of heterogeneous and homogeneous systems.

Fast evaluation of the adsorption energy of organic molecules on metals via graph neural networks

Modeling in heterogeneous catalysis requires the extensive evaluation of the energy of molecules adsorbed on surfaces. This is done via density functional theory but for large organic molecules it requires enormous computational time, compromising the viability of the approach. Here we present GAME-Net, a graph neural network to quickly evaluate the adsorption energy. GAME-Net is trained on a well-balanced chemically diverse dataset with C1-4 molecules with functional groups including N, O, S and C6-10 aromatic rings. The model yields a mean absolute error of 0.18 eV on the test set and is 6 orders of magnitude faster than density functional theory. Applied to biomass and plastics (up to 30 heteroatoms), adsorption energies are predicted with a mean absolute error of 0.016 eV per atom. The framework represents a tool for the fast screening of catalytic materials, particularly for systems that cannot be simulated by traditional methods.

Catlas: an automated framework for catalyst discovery demonstrated for direct syngas conversion

Catlas may be used with off-the-shelf pretrained models to explore large design spaces for catalyst discovery and has been used here to identify promising materials for the direct conversion of syngas to multi-carbon oxygenates.