Abstract
We describe a modified version of AlphaFold2 that incorporates experiential distance distributions into the network architecture for protein structure prediction. Harnessing the OpenFold platform, we fine-tuned AlphaFold2 on a small number of structurally dissimilar proteins to explicitly model distance distributions between spin labels determined from Double Electron-Electron Resonance (DEER) spectroscopy. We demonstrate the performance of the modified AlphaFold2, referred to as DEERFold, in switching the predicted conformations guided by experimental or simulated distance distributions. Remarkably, the intrinsic performance of AlphaFold2 substantially reduces the number and the accuracy of the widths of the distributions needed to drive conformational selection thereby increasing the experimental throughput. The blueprint of DEERFold can be generalized to other experimental methods where distance constraints can be represented by distributions.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
↵* Email: hassane.mchaourab{at}vanderbilt.edu