Previous work has demonstrated that the relationship between sign

Previous work has demonstrated that the relationship between signal and noise correlations

can Selleck Wnt inhibitor dramatically affect population coding (Gu et al., 2011). Thus, we asked whether the observed differences in correlation structure influence the coding of motifs by CLM neurons. Because motif identity is best regarded as a nominal, or categorical, variable (i.e., motifs cannot be easily described with a small number of parameters), we use multinomial logistic regression (MNLR) to find the optimal classifier that maximizes the predictability of motif identity from the firing rates of multiple neurons (Long, 1997) (Experimental Procedures). This technique is particularly well suited to our data because it can be applied to any number of neurons and any number of nominal categories (i.e., motifs). Figure 5 depicts the optimal classifier for a single pair of neurons responding to task-relevant motifs and shows that the classification boundaries follow many of the firing rate patterns that distinguish the motifs. In the example case (Figure 5), the classifier correctly predicted motif identity with 51% accuracy (far better than chance performance of 25%). The MNLR

model provides a rigorous quantification of the ability of CLM neurons to discriminate between different motifs. Using the MNLR model, we first asked whether the relationship between signal and noise correlations benefitted motif discrimination Cabozantinib supplier performance. We expressed this potential benefit as the “classification ratio,” which is simply the ratio of the MNLR classification accuracy with correlations intact to the classification accuracy without correlations (i.e., with trials shuffled, Experimental Procedures). Classification ratios greater than one indicate that correlations improve encoding while classification ratios less than one indicate that correlations impair

encoding. Consistent with theoretical predictions (Averbeck et al., 2006; Gu et al., 2011), Electron transport chain we find that the effect of correlation on encoding depends strongly on the relationship between signal and noise correlations (Figure 6A). For neuron pairs with positive noise correlations, the classification ratio is larger for pairs with negative signal correlations than for pairs with positive signal correlations (t test, p = 2.2 × 10−10; Figure 6A). Conversely, for neuron pairs with negative noise correlations, the classification ratio is larger for pairs with positive signal correlations than for pairs with negative signal correlations (t test, p = 0.044; Figure 6A). Thus, our data demonstrate that the observed correlations improve encoding when signal and noise correlations are of opposite sign and impair encoding when signal and noise correlations are of the same sign (two-way ANOVA interaction term, p = 1.8 × 10−8).

Leave a Reply

Your email address will not be published. Required fields are marked *

*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>