Reversing Cellular Senescence and Extracellular Matrix Degeneration: A Multiscale Model of Senolytic Intervention and Verifiable Recovery Predictions

A companion framework recently cast the coupling between cellular senescence and extracellular-matrix (ECM) degeneration as a single multiscale object, with a bistable cellular switch drivinga self-propagating tissue-scale wave of matrix collapse. If that forward cascade is the disease, thenatural question is whether—and when—it can be run in reverse. Here we extend the same mech-anistic backbone with an intervention layer and ask a quantitative therapeutic question: underwhat conditions does senolytic clearance of senescent cells, alone or combined with a matrix-restorative agent, not merely halt but reverse tissue degeneration? We report five computationalresults, each a real simulation of the coupled system. (i) A pharmacokinetic/pharmacodynamic(PK/PD) model shows that a single senolytic course flips the cellular state from the senescentfixed point (s∗ = 0.651) back to the healthy one (s∗ = 0.125), but only above a sharp criti-cal dose (D50 ≈ 0.97): reversal is all-or-nothing. (ii) Sustained clearance shifts the underlyingsaddle-node bifurcation, and above a critical clearance rate κc = 0.044 the senescent state ceasesto exist altogether—a cure in the dynamical sense. (iii) In the spatial model, treatment arreststhe invasion front and drives it into full retreat, with mean ECM stiffness recovering from 0.74back toward baseline while an untreated control saturates at complete degeneration. (iv) Matrixrecovery is time-critical: because degraded matrix scars irreversibly below a stiffness threshold,a therapeutic window exists (here closing at t ≈ 224) beyond which the cellular switch is stillreversible but the matrix is not. (v) A senolytic combined with a matrix-restorative agentis synergistic by the Bliss criterion and dose-sparing, lowering the senolytic dose needed fora target outcome by roughly one third. Finally, we train a deep ensemble of heteroscedasticregressors that predicts final tissue health from treatment parameters with calibrated uncer-tainty (expected calibration error reduced from 0.081 to 0.006 by post-hoc recalibration, pointaccuracy R2 = 0.86), turning the framework’s qualitative claims into falsifiable, error-barredpredictions. All parameters are illustrative rather than fitted to a specific tissue; the contribu-tion is a transparent, reproducible mechanism-to-prediction pipeline for reversal, not a validatedclinical model