# (1) Source Data We present the source data for the figures on our main article "Virtual Node Graph Neural Network for Full Phonon Prediction". For figure 2-4, we include the input and output data for the three models: Vector Virtual Nodes (VVN), matrix Virtual Nodes (MVN), and Momentum-dependent Matrix Virtual Nodes ($k$-MVN). Figure 5 covers the comparison of computation time between our $k$-MVN and M3GNet. Also, as the Extended Data, we provide the case predicting $\Gamma$-phonon without applying virtual nodes (NOVGNN). # (2) $\Gamma$-Phonon Database Given the high-quality phonon prediction using the machine-learning approach, particularly the matrix virtual nodes (MVN) approach for complex materials, here we present a new phonon database containing the phonon spectra for the entire 146,323 Materials in Materials Project (MP) as of 2022 computed by the MVN approach. Given the importance of zone-center $\Gamma$-phonons, which are measurable with more accessible equipment like Raman scattering, here we limit the database to $\Gamma$-phonons and will leave the full phonon database for future works. This new $\Gamma$-phonon database offers the possibility to analyze the lattice dynamics for many compounds. It could be used as a useful tool to be compared with Raman scattering results for crystalline materials. The whole database is stored as a dictionary, whose keys are the MP ID number. The dictionary values for each material are also dictionaries, which contain basic information: the number of atoms per unit cell, chemical formula, space group, crystal structure and phonon at $\Gamma$-point ($\text{cm}^{-1}$). # (3) Zeolite Phonon Prediction We performed phonon band prediction of zeolite materials, which are usually hard to compute phonon in existing methods. We provide theb input and outputs for the phonon band prediction. # (4) Anharmonic Phonon Calculation We also present the dataset of anharmonic phonon calculation. We keep 131 materials with 2 atoms per unit cell in 'anharmonic_fc_2atms.pkl'. The data is stored as a Pandas dataframe. For each materials with the Materials Project ID, you can find the data below (following parennthesis represents the shapes: 'mesh' (3,), 'frequency' (220, 6), 'gamma' (101, 220, 6), 'qpoint' (220, 3), 'temperature' (101,), 'grid_point' (220,), 'group_velocity' (220, 6, 3), 'gv_by_gv' (220, 6, 6), 'heat_capacity' (101, 220, 6), 'kappa' (101, 6), 'mode_kappa' (101, 220, 6, 6), 'weight' (220,), 'kappa_unit_conversion' (1,), 'fc3' (16, 16, 16, 3, 3, 3), 'p2s_map_fc3' (2,) , 'fc2' (16, 16, 3, 3), 'p2s_map_fc2' (2,), 'physical_unit_fc2' (1,), 'gruneisen' (220, 6), 'gruneisen_tensor' (220, 6, 3, 3).