(A) The network model configuration for the reconstructed SAI afferent in Figure 1D’. (B) Data flow through computational models and example outputs from each module: a finite element model (FEM) produces strain energy density (SED) at transduction units, transduction functions (Trans <# merkel cell–neurite complexes>) predict transduction currents (I(t)) and a leaky integrate-and-fire (LIF) array produces spike times. (C) The model’s predicted spike-timing variability, assessed by the distribution of normalized ISIs during static-phase responses (black bars: N = 1,591 intervals), corresponded to the skewed Gaussian distribution previously reported for mouse SAI afferents (orange bars: N = 3,348 intervals from 11 afferents; Wellnitz et al., 2010). To compare ISIs across a range of displacement magnitudes, each ISI was normalized to the mean interval for its stimulus. (D) Simulated firing rates (black symbols) from the model configuration in A were fitted to linear regressions of ramp-phase (blue dotted line: ramp acceleration = 20 mm·s−2, pink dotted line: ramp acceleration = 1143 mm·s−2) and static (orange dotted line) responses pooled from the SAI afferents shown in Figure 3B. Goodness of fit = 0.96 (fractional sum of squares). (E) Displacement–response relations from models configured with different primary cluster sizes. All configurations had 17 total transduction units and four spike initiation zones. Mean firing rates during the static phase of displacement are plotted (mean ± SD, N = 15 simulations per displacement). Displacement-response curves were compared by fitting with exponential regressions (R2 ≥ 0.99). Increasing or decreasing primary cluster size by two transduction units significantly changed the best fits (8 vs 10: p=0.004; 6 vs 8: p=0.017, extra sum-of-squares F test). Legend indicates the distribution of transduction units at spike initiation zones.