Solid binding peptides (SBPs) are a versatile class of engineered macromolecules with a huge range of applications including biomimetic mineralization, shape-selective nanoparticle synthesis, biomedical coatings, stimulus responsive particle assembly and more. Their tremendous power and diversity of application stems from the unique properties of biomolecules and the enormous phase space available to intrinsically disordered peptides. However, this massive application space is a double-edged sword as the properties of SBPs arise from the overlapping features of the sequence, surface and environmental conditions. Further, experimental probes of the structure and dynamics of nano-bio interfaces involving SBPs are costly, slow and extremely difficult to perform at scales required for phenomenological modeling.
Physics-based modeling tools such as molecular dynamics (MD) simulations are an important complement to experiments. MD simulations can predict important thermodynamic and kinetic quantities that reveal mechanisms of binding and help identify sequence-structure-energy relationships. However, the application of MD simulations is computationally demanding and requires significant expert knowledge, which can blunt the limit of these approaches. These limitations naturally raise questions of if and how data driven tools based like machine learning (ML) could be used to augment the limitations of MD and provide practical solutions to the challenge of SBP design.
This seminar will provide an overview of the three areas of the Pfaendtner research group’s efforts in application of ML/AI to SBP design. First, I will discuss the fundamentals of molecular data science. Second, through the lens of a data driven molecular optimization scheme, I will highlight contributions our group has made in the area of physics-based modeling of SBP/surface interactions. Finally, I will describe how this comes to together in recent projects that leverage high throughput simulations and data driven modeling for SBP design and discovery.