Fig. 6: Generation of polymer topologies with target \(\langle {R}_{{{{\rm{g}}}}}^{2}\rangle\).
From: Property-guided generation of complex polymer topologies using variational autoencoders
Topologies are generated aiming for target \(\langle {R}_{{{{\rm{g}}}}}^{2}\rangle\) values of 7.5 ± 2, 30 ± 2, and 50 ± 2. Each generated topology is accompanied by its type and the predicted \(\langle {R}_{{{{\rm{g}}}}}^{2}\rangle\) from TopoGNN, presented in parentheses on the x-axis. A violin plot showcases the revalidation of \(\langle {R}_{{{{\rm{g}}}}}^{2}\rangle\) via MD simulation for every topology. The gold dot marks the \(\langle {R}_{{{{\rm{g}}}}}^{2}\rangle\), while the white dot stands for the median. The black bar represents the interquartile range, and the plot width reflects the distribution density of \({R}_{{{{\rm{g}}}}}^{2}\). Two dashed lines highlight the \(\langle {R}_{{{{\rm{g}}}}}^{2}\rangle\) range used in the guided search.
