Linear MD encoders outperform nonlinear encoders when extrapolating to unseen regions of protein conformational space

Summary

Linear encoders outperform nonlinear encoders when extrapolating to unseen regions of protein conformational space (1). This was observed in MD simulations using machine learning to learn a reduced representation of the conformational space.

See also

• Protein folding neural networks can be combined with VAEs for conformational modeling

1.

Vani BP, Aranganathan A, Wang D, Tiwary P. AlphaFold2-RAVE: From Sequence to Boltzmann Ranking. Journal of Chemical Theory and Computation. 2023;19(14):4351–4. Available from: https://doi.org/10.1021/acs.jctc.3c00290