Fig 1

The HPNN architecture for (a) PES evaluation (generalized global neural network) and (b) structure generation (diffusion generative models). The figures are extracted from works by Yang et al. [31] and Guo et al. [46] respectively."

Fig 2

The architecture and user workflow of LASPAI website. The blue and green dotted box shows the architecture of the backend and frontend, respectively."

Fig 3

The generation of different polymorphs of TiO2 using the solid generation page. (a) The list of generated TiO2 structures sorted by unit energy. (b) and (c) are the first and second entries in (a), corresponding to the generated and optimized structures of anatase and rutile, respectively. The red and grey spheres represent O and Ti atoms, respectively."

Fig 4

The adsorption of pyridine on γ-Al2O3. (a) The workflow of adsorption study in LASPAI. (b) The minimum γ-Al2O3 (γ-AD) structure found by Yang et al. [74]. (c) The (512) facet of γ-AD structure, corresponding to (110) facet of γ-Al2O3. (d) The generated adsorption structure of pyridine on Al2O3. The white, grey, blue, red and pink spheres represent H, C, N, O and Al atoms, respectively."

Fig 5

Atomic simulation of oil/water phase separation. (a) The workflow for performing phase separation simulation on LASPAI. (b) The generated 1:1 molecular crystal of water and CS2. (c) The 1 : 1 mixed starting solution structure. (d) The structure after MD simulation with the two phases separated. The white, grey, red and yellow spheres represent H, C, O and S atoms, respectively."

Fig 6

TS structure identification of esterification reaction. (a) The workflow for identifying TS structure on LASPAI. (b) The molecular graph used to specify the reaction where the reacting bond is represented as dotted line. (c) The generated TS structure. (d) The calculated reaction energy profile by GGNN potential and DFT, where the TS structure is shown in inset."