Predictive neural networks accelerate identification of mechanistically validated small-molecule modulators of TLR7...

Toll-like receptor 7 (TLR7) is a key innate immune sensor implicated in autoimmune and inflammatory disorders. We report the discovery of novel small-molecule TLR7 antagonists, RTin7 and RTin11, using an AI-guided workflow combining a deep neural network (SMILES2ActNet), in silico screening, and medicinal chemistry optimization. The neural network accurately prioritized biologically active candidates from a large virtual chemical library. Both compounds exhibited low cytotoxicity and selectively inhibited imiquimod-induced proinflammatory cytokine release (TNF-α, IL-6, IL-8), with RTin11 showing superior potency in the low micromolar range compared to RTin7. Mechanistic studies demonstrated that RTin11 acts as a competitive antagonist at the TLR7-ligand interface, while RTin7 exhibits a distinct non-competitive inhibitory profile, with both compounds suppressing NF-κB and MAPK signaling. Molecular dynamics and MM/PBSA analyses revealed that RTin11 promotes receptor stabilization via cooperative structural adaptation, whereas RTin7 allows moderate flexibility, highlighting distinct binding behaviors. This study demonstrates the effectiveness of integrating deep learning with experimental validation for identifying selective, mechanistically validated TLR7 inhibitors as candidates for further therapeutic development. This integrative AI-to-experiment workflow may serve as a generalized model for identifying small-molecule modulators of pattern recognition receptors.