Project Summary/Abstract Nearly all modern drug discovery efforts begin with a virtual screening campaign, and there are now several drugs available on the market that were originally identified through computational efforts. Small molecule drug discovery, in particular, has benefited from the integration of computational chemistry tools that have become capable of the necessary throughput due to advances in computational power and algorithmic development. Binding free energy calculations have enhanced the practice of small molecule lead optimization and have helped expedite the development of effective drugs to transform the healthcare landscape. Peptide drugs are widely considered to be a promising, yet currently underutilized, class of drugs due to their high level of tunability with modern synthesis methods. Their large exposed surface area relative to small molecule drugs supports an increased level of specificity and the roughly 80 peptide drugs currently available on the market represent some of the highest therapeutic indices of known drugs. Peptide drugs even offer the opportunity to address targets that were previously thought to be undruggable. Peptide drugs are situated to similarly benefit from computational lead optimization strategies, but the application of existing tools for accelerating discovery efforts is less established than for small molecule drugs. The barrier for achieving widespread success in computational peptide lead optimization is two-fold: the increased size of peptide drugs makes the necessary conformational sampling more challenging, and the diversity in chemical character among different peptide drugs makes it difficult to fit a reliable classical simulation model. The former challenge can be overcome through efficient conformational sampling approaches, while the latter can be addressed by quantum mechanical calculations that avoid the need for a restrictive fit to analytical models. Overcoming both of these challenges simultaneously requires an innovative approach, because the solutions to each of them independently have been, until recently, mutually exclusive. The objective of the proposed research is to make computational lead optimization reliable for peptide drug discovery by developing a machine learning-based peptide simulation model that simultaneously overcomes both of the aforementioned challenges. We will achieve this by training a machine learning potential on the basis of quantum mechanical simulation data, which will be implemented on our QSP Life molecular simulation platform for efficiently calculating peptide binding free energies.