Membrane proteins are predicted by PLMs via solvent-exposed hydrophobic residues

Summary

Attention matrices of PLMs can capture whether a protein sequence is interpreted as a soluble protein or membrane protein (1). In membrane proteins, more attention is given to solvent-exposed hydrophobic residues.

See also

• Inverse folding models can distinguish between and design backbones belonging to transmembrane and soluble beta-barrels

1.

Wenzel M, Grüner E, Strodthoff N. Insights into the inner workings of transformer models for protein function prediction. Bioinformatics. 2024;40(3). Available from: https://doi.org/10.1093/bioinformatics/btae031