Student Researchers: Evan Stull , Mia Li , Dr. Gregory Wiedman

Abstract

The ever-growing scope of scientific computing, powered by large technical improvements in processing power and artificial intelligence, allows for tremendous advancements in drug discovery and pharmaceutical research. Using AI diffusion models such as RFdiffusion, it is possible to take known structures, such as those of enzymes like 9OMV, a transmembrane lipid flippase whose function is essential to the operation of fungal cells, and, using large datasets of known binding affinities and biological patterns, generate new binding peptides. These candidate peptides can be designed to target known inhibition sites in order to kill the cell, and are validated through the use of computational peptide folding programs such as AlphaFold3 and AutoDock-GPU. Candidates which show a strong binding affinity and low predicted inhibitory concentrations can then be validated through manual synthesis and inhibition assays with the targeted microbe. The use of large computational resources allows the process of screening thousands of candidates before reaching the manual stage to take orders of magnitude less time than the traditional methods of de novo synthesis, and this continues to shrink as time goes on.