3D CNN-CVAE and LSTM-guided de novo design of rotigotine derivatives targeting dopamine D1 and D3 receptors in restless legs syndrome.

Restless legs syndrome (RLS) is a chronic neurological disorder associated with altered dopaminergic signaling, while long-term dopamine agonist therapy may be limited by augmentation and reduced clinical benefit. This study applied a three-dimensional convolutional neural network conditional variational autoencoder integrated with long short-term memory molecular generation to design rotigotine-derived candidates targeting dopamine D1 and D3 receptors. Fourteen unique derivatives were generated and screened through pocket prediction, molecular docking, interaction profiling, pharmacophore analysis, density functional theory, ADMET prediction, 500 ns molecular dynamics simulations performed in explicit water without a lipid bilayer, and MM/GBSA calculations. Docking protocol validation using DockRMSD yielded RMSD values of 1.009 Å for Mevidalen in the D1 receptor and 1.359 Å for PD-128907 in the D3 receptor, supporting acceptable reproduction of both experimental binding poses. In docking-based ranking, AI Derivative 1 was placed above rotigotine for D1 (-5.744 vs -5.629 kcal/mol) and D3 (-7.934 vs -7.288 kcal/mol) and formed additional hydrogen-bond and hydrophobic contacts; the D1 difference of approximately 0.1 kcal/mol lies within the intrinsic uncertainty of docking scoring functions and is not interpreted as evidence of improved affinity. Docking scores are used throughout as ranking and prioritization metrics rather than quantitative binding free energies. It also exhibited a smaller HOMO-LUMO gap, a higher predicted LD50, and a shift from toxicity class 3 to class 4. Molecular dynamics analyses, including RMSD, RMSF, radius of gyration, solvent-accessible surface area, hydrogen bonding, DCCM, PCA, and free-energy landscapes, are reported as descriptive observations from single non-membrane trajectories and are not used to claim physiological relevance or superior binding stability. MM/GBSA estimates were favorable under the applied computational conditions. AI Derivative 1 therefore represents a computationally prioritized lead requiring experimental validation.