Fig. 9

Multidimensional Boltzmann sampling applied to RNA design. For the example of Fig. 5, we target G C content 85% and respective energies E1=-40, E2=-40, E3=-30 for the target structures (with tolerances of 5% GC content and 0.5 kcal/mol energy). Infrared ’s multidimensional Boltzmann sampling (MDBS) strategy starts from uniform sampling (weights 0 for every feature). It iteratively generates Boltzmann samples and updates the weights to move the (estimated) expectation closer to the targets. A Accepted samples as well as root mean square distance (RMSD) to the targets during this procedure, which considered over 70,000 total samples to generate 100 targeted samples. B Kernel density estimate plots: distributions of features for uniform sampling (blue) and sampling at the end of the MDBS run (red), where distributions are shifted to the targets (dashed red lines)