(342g) Inverse Design of Molecular Probes to Bind with Water Contaminants

Traditional water treatment systems are not designed to address many of the emerging classes of contaminants appearing in waterways and wastewater streams. Each water source has a unique mixture of solutes with varying degrees of environmental health and safety concern, and treating to remove them all is unlikely to be the most efficient strategy. Rather, it is desirable to design treatment methods capable of targeted, selective removal. To address the challenge of recalcitrant water pollutants, we take assistance from machine learning (ML) for navigating the large sequence and structure space of molecular probes by determining the key features that can enhance the selectivity and binding of targeted species. We describe our development of a platform combining deep representational learning, surrogate model construction, and Bayesian optimization to rationally traverse molecular design space and identify probes with high binding affinity and selectivity. We couple high throughput virtual screening using enhanced sampling all-atom molecular simulations with targeted experimental synthesis and testing within an integrated computational/experimental screening protocol. The optimal probes discovered by this approach will ultimately be used for detection and removal of organic pollutants from water.