Protein language models and PLM-based structure prediction generalize to de novo designed proteins

Summary

PLMs and PLM-based structure predictors generalize to some de novo designed proteins (1). Synthetic proteins designed using ESM were successfully expressed in the wet lab and adopted a structure consisted with predicted contacts. Likewise, ESMFold correctly predicted the structures of other de novo designed proteins (such as those designed by (2); see (3)).

Details

They use Gibbs sampling to generate new protein backbones by jointly considering $p(structure|sequence)$ and $p(sequence)$ (see below).

Figures

Ref (1)

See also

• Inverse folding from MD simulations

1.

Verkuil R, Kabeli O, Du Y, Wicky BIM, Milles LF, Dauparas J, et al. Language models generalize beyond natural proteins. openRxiv; 2022. Available from: https://doi.org/10.1101/2022.12.21.521521

2.

Praetorius F, Leung PJY, Tessmer MH, Broerman A, Demakis C, Dishman AF, et al. Design of stimulus-responsive two-state hinge proteins. Science. 2023;381(6659):754–60. Available from: https://doi.org/10.1126/science.adg7731

3.

del Alamo D, Jeliazkov JR, Truan D, Karpiak JD. Conformational sampling and interpolation using language-based protein folding neural networks. openRxiv; 2023. Available from: https://doi.org/10.1101/2023.12.16.571997