From affinity to kinetics: SPR as the analytical backbone of AI-driven drug discovery

초록

AI-driven drug discovery can now propose candidates faster than experimental triage can validate. However, many models still rely on static affinity or potency endpoints such as KD, IC50, or pEC50. This reliance creates a kinetic blind spot, as pharmacological durability often depends on dissociation kinetics and residence time. Surface plasmon resonance (SPR) can transform computational binding hypotheses into reusable, time-resolved learning signals. Here, we define the affinity-to-kinetics gap, link residence time to functional persistence, and analyze two AI-enabled workflows. In PRMT5 library compression, screening reduced 460,000 compounds to a single confirmed SPR binder (KD = 21.5 mu M); in TNIK generative design, INS018_055 showed potent SPR affinity (KD = 4.32 nM). Across these cases, SPR functions as a quality-control filter and experimental binding validator rather than a terminal confirmation assay. We identify five requirements for AI-ready SPR data: raw sensorgram retention, consistent fitting, explicit QC labels, experimental provenance, and reusable data schemas. Treating sensorgrams, fit diagnostics, and QC labels as data infrastructure facilitates AI-SPR co-learning, providing a practical route to kinetic-aware, auditable, and reusable decision systems.

키워드