Fig. 3: Neurons represent sensory variables, latent world states, and motor variables.
From: Dynamical latent state computation in the male macaque posterior parietal cortex
a Causal structure of the task, illustrating the recurrent nature of the interaction between sensory inputs, latent states, and motor outputs. b Schematic of the generalized additive model (GAM) used to fit spike trains of single neurons. c Activity of simultaneously recorded neurons during a random thirty second epoch during the experiment (left) and the corresponding prediction reconstructed using the model (right). Neurons are arranged according to their contribution to the leading principal component (PC) (bottom—lowest; top—highest) for visualization. d Proportion of neurons tuned to different variables. Error bars denote ± 1 standard error of binomial proportions (n = 244). e Cumulative distribution of the contribution of different predictors, calculated as the reduction in variance explained by the model after removing those predictors. Black shows the distribution of the variance explained by the full model. f Top: Example tuning functions showing sensitivity of neurons to different task variables and other explanatory variables. Shaded regions denote ± 1 SEM across validation sets (n = 10). Bottom: Peak-normalized tuning functions of all significantly tuned neurons, sorted according to the peak feature. Across the population, tuning to individual task variables had peak responses that tiled all values of the feature space. In contrast, most neurons produced stereotyped responses to Local field potential (LFP) phase. Source data are provided as a Source data file.
