Fig. 1: The graph neural network (GNN) design and its outputs. | npj Computational Materials

Fig. 1: The graph neural network (GNN) design and its outputs.

From: Unified physio-thermodynamic descriptors via learned CO2 adsorption properties in metal-organic frameworks

Fig. 1

a The detailed IsothermNet architecture, composed of graph attention (GAT) and graph convolutions (CGCNN). The predictive capabilities of IsothermNet as revealed by the parity plots for (b) uptake and (c) heat of adsorption at 0.5, 5, and 50 bars. The 50% (orange contour), 80% (brown contour), and 95% (black contour) confidence intervals are marked. The color gradient and marginal kernel density estimation (KDE) plots denote the density distribution/point concentration along the predicted (from IsothermNet) and target (ground truth from GCMC simulations) axes.

Back to article page