Recent breakthroughs in deep learning—highlighted by the 2024 Nobel Prize in Chemistry—have transformed our ability to study and design protein systems. Models like AlphaFold can predict complex 3D protein structures directly from sequence data. More impressively, modern co-folding models can also predict which small-molecule ligands bind to which specific proteins and how. These architectures are primarily employed in drug discovery to help identify candidates to restore protein function or inhibit disease.
Of course, drugs are not the only type of molecules that can bind to proteins and affect their function. Hazardous nano-pollutants can interact with proteins via similar mechanisms. In particular, per- and polyfluoroalkyl substances (PFAS)—so-called “forever chemicals”—have been reported to disrupt protein function in multiple ways, e.g., accumulating in human tissue, impairing metabolic pathways, and damaging retinal function. While molecular dynamics simulations and wet-lab experiments can probe specific protein-PFAS pairs, screening the entire human proteome to pinpoint which protein motifs are most vulnerable remains a daunting challenge. Mapping these interactions is key to mitigating health risks, guiding regulatory policies, and designing safer chemical alternatives.
In this project, you will explore the potential of co-folding models to identify PFAS types and protein motifs that are most likely to bind, potentially affecting human health. You will benchmark existing co-folding models, validate them against molecular dynamics simulations and, possibly, fine-tune them. Experience with machine learning and/or protein simulation is desirable.
