SELECTIVE MECHANOPORATION OF CANCER MEMBRANES VIA ACOUSTIC RESONANCE AND LDH NANOPARTICLES: A CALIBRATED MULTISCALE IN SILICO STUDY

<jats:p>Targeted non-invasive physical oncology requires precise biomechanical differentiation between malignant and non-malignant cells [1, 3]. Here, we present a calibrated in silico multiscale workflow coupling multi-domain viscoelastic Finite Element Analysis (FEA) [21], Coarse-Grained Molecular Dynamics (CG-MD) [24], and surrogate Machine Learning (ML) [31]. We model breast adenocarcinoma cells (MCF-7) in COMSOL Multiphysics 6.1 with realistic biophysical parameters. A 3.0 mm(≥ 𝜆/4) Perfectly Matched Layer (PML) bounds the domain: nucleus (𝐸 = 3.5 kPa, 𝜂 = 4.5 Pa ⋅ s) [4, 5], cytoplasm (𝜂 = 2.5 Pa ⋅ s) [17], actin cortex (𝐸 = 2.2 kPa) [6], and bilayer membrane (𝐸 = 12.5 ± 2.1 kPa) [2, 7]. Under acoustic excitation (𝑃0 = 35 kPa), dynamic Fluid-Structure Interaction (FSI) frequency sweeps (20--150 kHz) predict a fundamental dipolar structural resonance mode for MCF-7 at 𝑓0 = 54.2 ± 15.2 kHz, serving as a testable in silico hypothesis for future experimental validation [33, 36]. Control non-malignant MCF-10A cells (𝐸mem = 18.5 kPa) exhibit peak response at 𝑓0 = 62.4 ± 12.1 kHz. Hexagonal platelet Layered Double Hydroxide (LDH) nanoparticles (𝑑 = 25 nm, 𝑡 = 4 nm, 𝜁-potential = +34.2 mV, 𝑍 = 13.2 MRayl) act as acoustic impedance-mismatched focalizers [11, 13]. These particles amplify localized boundary shear stress to 𝜏max = 14.8 ± 2.4 mN/m[14]. Converting FEA boundary stresses into lateral physical pull forces (𝐹lateral = 45 pN) within Martini 3 CG-MD simulations (15 × 15 × 12 nm3 box, 35.0 water/lipid, realistic lipidomics) across 𝑁 = 5independent trajectories yields dynamic pore nucleation at 𝑡pore = 142.5 ± 18.2 ns(95% CI) and equilibrium pore diameters of 𝑑 = 6.2 ± 0.8 nm[26, 34]. An XGBoost surrogate model evaluated with 5-fold cross-validation and 10% injected Gaussian noise achieves 𝑅2 = 0.814[31]. SHAP and SALib Sobol sensitivity analyses identify membrane Young's modulus (𝑆𝑇 = 0.61) and cytoplasm viscosity (𝑆𝑇 = 0.44) as primary sensitivity drivers [29, 32]. Corrected Pennes bioheat modeling predicts a maximum local thermal rise of Δ𝑇max = 0.72 ± 0.05∘C[19, 20], confirming a non-thermal mechanical mode of action [8].</jats:p>