Extended Data Fig. 10: Modeling pain behavior, pain state, and effort following formalin administration.

(a) The average pain input given when training and simulating the model, chosen as the sum of two Gaussians corresponding to phases 1 and 2 of the formalin response. Model input during training and simulation is a noisy version of this input. (b) The pain state of the model is an integral of the pain input passed through a rectified saturating nonlinearity, shown here. (c) Schematic of the neural network implementing the behavioral policy, also showing the addition of a third state dimension ‘NPY’ that is 1 in the presence of a competing survival need and 0 otherwise (used in “Modulation of Policy” simulations). (d) Example policy space after training one model in the “Modulation of Policy” scenario in which the model learns a distinct policy for the presence vs absence of a competing need state. The dark gray line indicates a trajectory in the pain-effort state space, while the light gray region indicates the portion of pain-effort state space in which the model produces a lick response (no lick response is produced outside this region). When NPY = 1, the lick region shifts upwards compared to NPY = 0, corresponding to an increase in the pain threshold required to produce a licking response. (e-f) Average ‘effort’ dynamics of 8 trained models for the baseline model (‘control’) vs a model that increases the effort cost of licking (e) or introduces a third axis to the behavioral control policy (f). Dark lines represent mean and lighter, shaded areas represent SEM. (g-o) Simulation results using alternative formulations of input integration, effort modulation, and policy modulation (see equation 2 for each manipulation in the Methods). (g-i) Simulated licking behavior (g), ‘pain’ state (h) and ‘effort’ state (i) in the baseline condition (‘control’) vs a competing need that alters integration of nociceptive input using manipulation 2. (j-l) Simulated licking behavior (j), ‘pain’ state (k) and ‘effort’ state (l) in the baseline condition (‘control’) vs a competing need that alters the effort cost of licking using manipulation 2. (m-o) Simulated licking behavior (m), ‘pain’ state (n) and ‘effort’ state (o) in the baseline condition (‘control’) vs a competing need that adds a third dimension to the behavioral control policy using manipulation 2. Data are expressed as mean ± SEM.

Source data