Machine learning and molecular design of self-assembling π-conjugated oligopeptides

Self-assembling oligopeptides present a means to fabricate biocompatible supramolecular aggregates with engineered electronic and optical functionality. We conducted molecular dynamics simulations of self-assembling synthetic oligopeptides with Asp-X cores that mediate hydrophobic stacking and electron delocalisation within the self-assembled nanostructure. The larger PDI cores elevated oligomerisation free energies by a factor of 2-3 relative to NDI and also improved alignment of the oligopeptides within the stack. Training of a quantitative structure-property relationship (QSPR) model revealed key physicochemical determinants of the oligomerisation free energies and produced a predictive model for the oligomerisation thermodynamics. Oligopeptides with moderate dimerisation and trimerisation free energies of produced aggregates with the best in-register parallel stacking, and we used this criterion within our QSPR model to perform high-throughput virtual screening to identify promising candidates for the spontaneous assembly of ordered nanoaggregates. We identified a small number of oligopeptide candidates for direct testing in large scale molecular simulations, and predict a novel chemistry DAVG-PDI-GVAD previously unstudied by experiment or simulation to produce well-aligned nanoaggregates expected to possess good optical and electronic functionality.