by Harald Feldmann
•
30 December 2023
Proteins like to fold into a state of lowest possible energy. In protein design, software is used to find the lowest state, but limitations in algorithms introduce artifacts where a so called 'local minimum' is selected rather than the absolute minimum. Imagine a lake in the mountains, thinking it is the lake at sea level. By making a map of the landscape you find all other lakes nearby which may also not be at sealevel, but you may also find the sea itself. A Machine Learning method that can map and optimize the energy landscape is subject of the publication 'Protein sequence design by conformational landscape optimization' in the 'Proceedings of the National Academy of Sciences', PNAS : https://www.pnas.org/doi/10.1073/pnas.2017228118 to which Harald Feldmann contributed (see supplemental data PDF).