Extended Data Fig. 1: High-dimensional, high-cost real-world tasks.
(a) Optimizing the mechanical properties of architected materials by DANTE and generative architecture design (GAD, baseline). (b) Uniform manifold approximation and projection (U-MAP) 2D representation of results from two methods. (c) Stimulated strain-stress curve of both methods. The inlet shows the density matrix. (d) Optimizing the electronic properties of complex concentrated alloys (CCAs) by DANTE and MCMC (baseline). (e) U-MAP 2D representation of input distribution from both methods. (f) The curves along a selected momentum path on the Fermi surface, a quantitative measure for describing the smearing. (g) Optimizing the protein-protein interactions (PPIs) using DANTE and other two methods. Box plots indicate median (middle line), 25th, 75th percentile (box), and 1.5 × interquartile range (whiskers). n = 41 for 1st and 2nd iterations, n = 42 for 3rd iteration, Native, GD (gradient descent), and MCMC. (h) An example of Alphafold2 predicted complex (pdbid: 4ib5). The cyclic peptide is designed by DANTE. (i) Interaction diagram of DANTE peptide with the target protein.
