Machine Learning‐Driven Discovery of GABA‐A Ligands with Pronounced Neuroprotective Efficacy and Resistance to P‐gp Efflux

Many studies have pointed to GABA-A receptors as a promising therapeutic target for promoting recovery after stroke, owing to their neuroprotective efficacy and the enhancement of synaptic GABAergic currents. However, identifying nanomolar-affinity ligands that evade P-glycoprotein-mediated efflux can pose a significant challenge. To overcome this barrier, we developed a structure-based machine learning workflow integrating molecular docking, which screened over 160 000 virtual analogs and identified eight synthetically accessible molecules. The synthesized compounds exhibited K_i values at the GABA-A receptor ranging from 50 to 1600 nM, with the most promising being 4d (K_i = 62 ± 11 nM) and 4h (K_i = 50 ± 3 nM), both of which confirmed PAM efficacy at GABA-A receptors. Both ligands exhibited neuroprotective activity by attenuating glutamate-induced Ca²⁺ overload, preserving mitochondrial membrane potential and enhancing cell viability following oxygen-glucose deprivation in HT-22 neurons. In MDR1-MDCKII bidirectional assays, compound 4d (azetidine derivative) exhibited symmetric permeability (efflux ratio = 0.94), while 4h (bicyclic amino alcohol) was identified as a P-gp substrate (efflux ratio = 2.04), suggesting that eliminating a single hydrogen-bond donor at the amide tail is critical for minimizing efflux. Collectively, this study identifies compound 4d as a potent, low-efflux GABA-A receptor PAM with neuroprotective properties, supporting its further investigation.