(15d) Multi-Fidelity Computational-Experimental Design of Self-Assembling ?-Conjugated Peptides

Using a hybrid multi-fidelity, multi-objective Bayesian optimization protocol we combine experimental characterization data with computational molecular dynamics simulations to design π-conjugated peptides capable of self-assembling into conductive nanowires. π-conjugated peptides represent a class of tailorable biocompatible molecules that self-assemble to form supramolecular networks supporting long-range charge transport with applications as peptide-based transistors and energy harvesting solar cells. The family of π-conjugated peptides considered in this work are composed of symmetric oligopeptide wings flanking a central π-core that have both experimentally and computationally been demonstrated to self-assemble into pseudo-1D nanoaggregates with emergent optoelectronic properties. Exhaustive consideration via either experiment or simulation of all realizable π-conjugated peptide candidates is prohibitively expensive due to the massive combinatorial molecular design space of all possible peptide wings. This motivated us to fuse approximate, high-volume, and easy-to-collect simulation data with expensive, low-volume, but accurate experimental data within a multi-fidelity Bayesian optimization platform to traverse our molecular design space and rationally discover candidate π-conjugated peptides with good self-assembly behavior. Molecules identified from our active learning search submit to experimental synthesis and characterization yield superior optoelectronic properties compared to previously known candidates, while data collected throughout our search also uncovers design rules linking optoelectronic functionality with molecular structure.