Can AlphaFold2 predict protein-peptide complex structures accurately?

A bstract In this preprint, we investigated whether AlphaFold2, AF2, can predict protein-peptide complex structures only with sequence information. We modeled the structures of 203 protein-peptide complexes from the PepBDB DB and 183 from the PepSet. The structures were modeling with concatenated sequences of receptors and peptides via poly-glycine linker. We found that for more than half of the test cases, AF2 predicted the bound structures of peptides with good accuracy, C α -RMSD of a peptide < 3.0 Å. For about 40% of cases, the peptide structures were modeled with an accuracy of C α -RMSD < 2.0 Å. Our benchmark results clearly show that AF2 has a great potential to be applied to various higher-order structure prediction tasks.