The widespread presence of persistent and toxic per- and polyfluoroalkyl substances (PFAS) in drinking water necessitates the development of sensitive and selective detection methods. Existing analytical techniques often suffer from high costs and complexity, while alternative sensing approaches like fluorescent and electrochemical sensors may lack the required performance in real water matrices. Field-effect transistor (FET) sensors offer a promising platform for rapid and portable PFAS detection, contingent on the availability of highly selective molecular probes. We have developed a Bayesian Optimization-driven computational high-throughput screening strategy to design functionalized β-cyclodextrin (CD) variants with significantly enhanced affinity and selectivity for perfluorooctanesulfonic acid (PFOS), a prevalent PFAS. Our approach involved defining a chemical design space of substituted CDs and employing biased molecular dynamics (parallel-biased metadynamics) simulations to efficiently estimate their binding free energies and selectivity against sodium dodecyl sulfate (SDS), a common interferent. Our computational screening efforts identified promising candidates with sub-nanomolar dissociation constants for PFOS (Kd ≈ 10-11 M) and a selectivity (KdSDS/ KdPFOS ≈ 5 x 104) approximately 5000 times greater than that of unmodified β-CD. Our results indicate that novel probe functionalized strategies are constructive in amplifying sensitivity and selectivity relative to unmodified β-CD. These findings provide a compelling direction for the development of highly sensitive and selective molecular probes for real-time PFAS detection using FET sensors, offering a potentially cost-effective and field-deployable solution. Furthermore, the interpretable design principles derived from this study can inform the development of probes for other environmentally relevant contaminants with similar chemical characteristics.
