Engineering conditional peptide self-assembly using AlphaFold

Programming self-assembly from peptide sequence remains difficult because existing design rules are only weakly predictive of emergent supramolecular behavior. Here, we show that AlphaFold Multimer can be used to engineer conditional peptide self-assembly by simultaneously selecting for co-assembly and against assembly of the individual components. Using model confidence metrics as proxies for assembly propensity, we designed peptide pairs that largely remain unstructured in isolation but form defined tape-like nanostructures upon mixing. Experimental measurements including circular dichroism, small-angle X-ray scattering, and cryo-electron microscopy validate the predicted behavior across multiple systems. These results establish artificial intelligence-guided design as a practical route to programmable supramolecular peptide materials.