Unbiased In-Silico Design of pH-Sensitive Tetrapeptides.

Oligopeptides are short peptides consisting of two to twenty amino acids (AA) that can spontaneously fold and assemble through a combination of hydrogen bonds and π-π interactions to form functional nanostructures. From an engineering perspective, oligopeptides folding and assembly can be designed to fabricate fibers, tubes, nanosheets, pallets, gels, vesicles, and nanoparticles, with applications in biomedicine, food science, regenerative medicine, and biosensing. Peptide-based biomaterials offer in fact the opportunity to combine the simplicity of small biomolecules with the functionality of proteins. The design of oligopeptides that assemble in nanostructured materials often follows principles of bioinspiration and rational design. Most of the short (i.e. &lt;5AA) oligopeptides-based biomaterials found in literature are directly derived from AA <i>motifs</i> with biological relevance (e.g. DFNKF, KLVFF) and, are composed of hydrophobic AA (e.g. FFF, VYV) to drive self-assembly in water. The combination of bioinspiration with rational design, despite successful, has limited the design of oligopeptides to few sequences tested experimentally, when compared to the <i>xⁿ</i> (where <i>x</i>=20 AA and <i>n</i>=number of AA in the oligopeptide) theoretical possibilities, and biased the AA choice to impart low solubility. Such restrictions have strongly limited the discovery of new AA sequences.<br/>As an alternative route, in silico tools can be used to predict oligopeptides’ properties, accelerating their design into biomaterials. Machine learning (ML) algorithms such as TorchMD, Convolutional neural networks, and Deep neural networks have been successfully used to quickly model and predict peptides’ folding, energies, and reaction pathways, but the limited accuracy and completeness of high-quality training data still limit the resolution of predictive results of ML when compared to molecular dynamics (MD) simulations, which in turn are extremely intensive. To combine the benefits and limit the individual weaknesses of MD and ML, hybrid ML-MD approaches are now pursued to accelerate simulations and improve the understanding of complex biomolecular systems (e.g. flexible molecular force fields, where ML tools are used to accelerate the simulation process). However, these tools are still in their infancy and their applications to decrease the intensity of MD simulations have to be fully explored.<br/>Here, we developed a new computational design protocol to discover oligopeptides that self-assemble in nanostructured biomaterials by combining an unbiased AA selection (i.e. agnostic to chemical features) with coarse grain (CG) MARTINI forcefield (highly parameterized for natural amino acids), which yields a speed-up of 2–3 orders of magnitude compared to atomistic forcefield. We focused on tetrapeptides as <i>n</i>=4 represents a wide but approachable sequences space (20⁴ possible unique sequences) to screen and test computational unbiased methods while possessing an amphiphilic form and the proven ability to self-assemble into nanofibrillar structures. This method allowed us to simulate the assembly of all the possible tetrapeptides without bias, resulting in the screening of appropriate side chain combinations to embed responsiveness to environmental stimuli, such as pH. This pH-triggered assembly of tetrapeptides can be used to engineer new biomaterials for drug delivery that assemble/disassemble in response to pH changes, nanofibrillar matrices for separation of large biomolecules like proteins, antimicrobial surfaces, and scaffolds to support for cell growth.