A Multilevel Mechanistic Platform for Antidepressant-Response Prediction Integrating Pharmacokinetics, Polypharmacology, Inflammation, Neuroplasticity, Large-Scale Networks and the Gut–Brain Axis: Chronological Development and Indicator-Level Evaluation Across 51 Drugs and 15 Drug–Nutraceutical Combinations

Background: Major depressive disorder (MDD) shows <50% remission even under optimized treatment, with response heterogeneity driven by polygenic, inflammatory, neurocircuit and exposome factors. Mechanistic prediction remains fragmented and often relies on arbitrary statistical weights violating biophysical constraints. Methods: We built a mechanistic, multilevel platform with no trainable statistical weights. The system couples seven subsystems: pharmacokinetics; multi-target pharmacodynamics; kynurenine metabolism; the GR/FKBP5 axis; BDNF-related neuroplasticity; large-scale networks; and MADRS turnover with adherence, photoperiod, and placebo terms. Constants were drawn from regulatory, database, and published sources. Indicator-level sample counts sum to N=844, but these observations do not form a single jointly measured cohort; synthetic data were used for stress testing rather than clinical validation. Results: Iterative development across seven chronological versions (V1->V7) resolved initial failures without weight tuning. Final V7 performance was as follows: PK MAPE 0.68% (R²=0.999, N=48); PD occupancy 0.55% (R²=0.999, N=48); kynurenine 0.36% (R²=1.00, N=52); GR/FKBP5 0.79% (R²=0.971, N=120); BDNF 3.61% (R²=0.139, N=110); networks 2.25% (R²=0.991, N=100); MADRS at week 8, 7.23% (R²=0.891, RMSE=1.19, N=300); synthetic drugs, 1.45% (R²=0.926, N=39); natural products, 3.08% (R²=0.260, N=12); and 15 composite combinations, 1.69% (R²=0.919). Across the 11 reported rows, mean MAPE was 2.16% and the maximum was 7.23%; AUROC was 0.88. Conclusions: A mechanistic platform can integrate processes across atomic-to-circuit scales and generate predictions for monotherapies and combinations while exposing limitations associated with bioavailability and blood–brain barrier penetration. Because no single cohort contains all biomarkers simultaneously, prospective collection of genomic, inflammatory, neuroimaging, epigenetic, adherence, and weekly 8-week MADRS data is required before clinical deployment. Transparency takes precedence over apparent performance.