Extended Data Fig. 6: An encoding model for reconstructing the activity of individual neurons using the continuously evolving experimental variables. | Nature Neuroscience

Extended Data Fig. 6: An encoding model for reconstructing the activity of individual neurons using the continuously evolving experimental variables.

From: Population coding of strategic variables during foraging in freely moving macaques

Extended Data Fig. 6

(a) A generalized linear model to reconstruct continuously evolving firing rates of individually recorded neurons. Rates are predicted from combinations of the task variables passed through a set of basis functions. The basis functions were pulse-shaped temporal delay filters for press, reward, and choice events. (The time of the reward and the choice events were assumed to match the time of the press after which the reward was delivered or the choice was made.) The basis functions for continuously evolving task variables (waiting time, reward ratio, and 2-dimensional ___location) were instantaneous power functions with powers of ½, 1, 2, 3, and 5. Altogether, 51 predictors were made using these 6 task variables. The model was fit to the training data using a Gaussian likelihood function; the trained model was used to reconstruct the neural activity for held-out testing data. (b) The improvement in the performance of the model when either one or both reward predictors were used alongside the other task variables in (a). Improvement was calculated as the percentage increase in the correlation between the recorded and reconstructed activities. While the waiting time improved the model performance for the entire population by 5% (\(p\)≪\({10}^{-3}\)), this improvement was insignificant for the reward ratio (p = 1) at the level of individually recorded neurons. (c) The reconstructed and recorded firing rates averaged across 6 neurons in a sample session. The neurons were selected from 60 simultaneously recorded neurons in this session by clustering them using the correlation matrix of their reconstructed activities (inset), then choosing all neurons in a sample cluster (bracket in the inset) to show here. (d) The correlation coefficient between each reward predictor and the reconstructed neural activity averaged across neurons in the same cluster vs. the number of neurons in the cluster. The vertical line shows a cluster size of 5, which was the lower bound of the cluster size for the clusters that were included in Fig. 4c. Across sessions, the average firing rates of clusters of ≥ 5 neurons were positively correlated with the waiting time (p = 0.004, top) and negatively correlated with the reward ratio (p = 0.004, bottom).

Back to article page