Limits...
Model-based analysis of pattern motion processing in mouse primary visual cortex.

Muir DR, Roth MM, Helmchen F, Kampa BM - Front Neural Circuits (2015)

Bottom Line: We also found a large proportion of cells that respond strongly to only one stimulus class.Our results show that a broad range of pattern integration processes already take place at the level of V1.This diversity of integration is consistent with processing of visual inputs by local sub-networks within V1 that are tuned to combinations of sensory features.

View Article: PubMed Central - PubMed

Affiliation: Laboratory of Neural Circuit Dynamics, Brain Research Institute, University of Zürich Zürich, Switzerland ; Biozentrum, University of Basel Basel, Switzerland.

ABSTRACT
Neurons in sensory areas of neocortex exhibit responses tuned to specific features of the environment. In visual cortex, information about features such as edges or textures with particular orientations must be integrated to recognize a visual scene or object. Connectivity studies in rodent cortex have revealed that neurons make specific connections within sub-networks sharing common input tuning. In principle, this sub-network architecture enables local cortical circuits to integrate sensory information. However, whether feature integration indeed occurs locally in rodent primary sensory areas has not been examined directly. We studied local integration of sensory features in primary visual cortex (V1) of the mouse by presenting drifting grating and plaid stimuli, while recording the activity of neuronal populations with two-photon calcium imaging. Using a Bayesian model-based analysis framework, we classified single-cell responses as being selective for either individual grating components or for moving plaid patterns. Rather than relying on trial-averaged responses, our model-based framework takes into account single-trial responses and can easily be extended to consider any number of arbitrary predictive models. Our analysis method was able to successfully classify significantly more responses than traditional partial correlation (PC) analysis, and provides a rigorous statistical framework to rank any number of models and reject poorly performing models. We also found a large proportion of cells that respond strongly to only one stimulus class. In addition, a quarter of selectively responding neurons had more complex responses that could not be explained by any simple integration model. Our results show that a broad range of pattern integration processes already take place at the level of V1. This diversity of integration is consistent with processing of visual inputs by local sub-networks within V1 that are tuned to combinations of sensory features.

No MeSH data available.


Model-based classification performs better than partial correlation (PC) analysis. (A) The distribution of Z-scored partial correlations for the component (Zc) and pattern (Zp) models is plotted, along with the decision boundaries used to classify responses under the PC analysis (dashed lines; see Materials and Methods). Each cell is plotted with a color and symbol according to its classification under our model-based analysis method (see text; Materials and Methods). Many of the neurons left unclassified by the PC analysis are successfully classified by our model-based approach [see cells plotted in the “Unclassified (PC)” region]. In addition, some cells classified by the PC analysis are classified differently, or left unclassified, by our model-based approach. (B) Cells classified by our model-based approach (gray) are just as well-predicted as cells classified by PC analysis (black; median LL −32 vs. −32, p = 0.88, rank-sum test, nMB = 355, nPC = 85). (C) However, cells left unclassified by the PC analysis could have been reasonably predicted by one of the two models (black; high LL). In contrast, cells left unclassified by our model-based approach were much more poorly predicted by one of the two models (gray; lower LL); this difference between the two classification approaches was statistically significant (median LL -36 vs. -32, p < 0.001, rank-sum test, nMB = 177, nPC = 244). Example traces c7–c9 illustrate cells which are differently classified by PC and our model-based approach. Gray traces indicate single-trial calcium responses; the single-trial responses used in our analysis are indicated as black dots. Blue and red curves indicate predicted single-trial distributions under the component and pattern models, respectively, generated as part of our model-based analysis. Insets in c7–c9 indicate both the model-based and PC classification of the corresponding cell. Labels in (A–C) indicate the measures corresponding to each example cell. Trace c7 shows a cell which is classified as a “pattern” cell by PC analysis, but “component” cell by our model-based analysis. The observed single-trial responses to plaid stimuli (black dots) are more likely under the component model than the pattern model (blue vs. red curves). Trace c8 shows a cell that could not be classified under PC analysis, but was classified as a component cell by our model-based analysis. Trace c9 shows a cell that was classified as a pattern cell by PC analysis, but was left unclassified by our analysis since the single-trial responses to plaid stimuli were equally well predicted by the component and pattern models (blue vs. red curves). Scale bars: 10 s and 10% ΔF/F. ***p < 0.001.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4525018&req=5

Figure 3: Model-based classification performs better than partial correlation (PC) analysis. (A) The distribution of Z-scored partial correlations for the component (Zc) and pattern (Zp) models is plotted, along with the decision boundaries used to classify responses under the PC analysis (dashed lines; see Materials and Methods). Each cell is plotted with a color and symbol according to its classification under our model-based analysis method (see text; Materials and Methods). Many of the neurons left unclassified by the PC analysis are successfully classified by our model-based approach [see cells plotted in the “Unclassified (PC)” region]. In addition, some cells classified by the PC analysis are classified differently, or left unclassified, by our model-based approach. (B) Cells classified by our model-based approach (gray) are just as well-predicted as cells classified by PC analysis (black; median LL −32 vs. −32, p = 0.88, rank-sum test, nMB = 355, nPC = 85). (C) However, cells left unclassified by the PC analysis could have been reasonably predicted by one of the two models (black; high LL). In contrast, cells left unclassified by our model-based approach were much more poorly predicted by one of the two models (gray; lower LL); this difference between the two classification approaches was statistically significant (median LL -36 vs. -32, p < 0.001, rank-sum test, nMB = 177, nPC = 244). Example traces c7–c9 illustrate cells which are differently classified by PC and our model-based approach. Gray traces indicate single-trial calcium responses; the single-trial responses used in our analysis are indicated as black dots. Blue and red curves indicate predicted single-trial distributions under the component and pattern models, respectively, generated as part of our model-based analysis. Insets in c7–c9 indicate both the model-based and PC classification of the corresponding cell. Labels in (A–C) indicate the measures corresponding to each example cell. Trace c7 shows a cell which is classified as a “pattern” cell by PC analysis, but “component” cell by our model-based analysis. The observed single-trial responses to plaid stimuli (black dots) are more likely under the component model than the pattern model (blue vs. red curves). Trace c8 shows a cell that could not be classified under PC analysis, but was classified as a component cell by our model-based analysis. Trace c9 shows a cell that was classified as a pattern cell by PC analysis, but was left unclassified by our analysis since the single-trial responses to plaid stimuli were equally well predicted by the component and pattern models (blue vs. red curves). Scale bars: 10 s and 10% ΔF/F. ***p < 0.001.

Mentions: Our model-based analysis method offers several advantages over classification of pattern and component cells using PC analysis. Firstly, we are able to classify a greater proportion of responses into pattern and component classes (304 vs. 85 cells; Figure 3A), without loss of performance (no difference in log likelihoods; Figure 3B). Model predictions were also significantly better correlated with the recorded responses of successfully classified neurons than with responses of unclassified neurons (Supplementary Figure 1; median correlations 0.55 vs. 0.08; p < 0.01, rank-sum test). For some cells, the trial-averaged response used by PC analysis was a poor description of the cell's full response, so that our model-based approach assigned a different category than PC analysis (e.g., Figure 3 trace c7). The cells left unclassified by our model-based method had responses that were poorly explained by either of the pattern or component models (low log likelihoods; Figure 3C), or were equally well explained by both models (see Figure 3 trace c9). However, cells left unclassified under the PC analysis in general had responses that fit reasonably well into one of the pattern or component models (significantly higher log likelihoods, p < 0.001; Figure 3C; see Figure 1 trace c2 and Figure 3 trace c8).


Model-based analysis of pattern motion processing in mouse primary visual cortex.

Muir DR, Roth MM, Helmchen F, Kampa BM - Front Neural Circuits (2015)

Model-based classification performs better than partial correlation (PC) analysis. (A) The distribution of Z-scored partial correlations for the component (Zc) and pattern (Zp) models is plotted, along with the decision boundaries used to classify responses under the PC analysis (dashed lines; see Materials and Methods). Each cell is plotted with a color and symbol according to its classification under our model-based analysis method (see text; Materials and Methods). Many of the neurons left unclassified by the PC analysis are successfully classified by our model-based approach [see cells plotted in the “Unclassified (PC)” region]. In addition, some cells classified by the PC analysis are classified differently, or left unclassified, by our model-based approach. (B) Cells classified by our model-based approach (gray) are just as well-predicted as cells classified by PC analysis (black; median LL −32 vs. −32, p = 0.88, rank-sum test, nMB = 355, nPC = 85). (C) However, cells left unclassified by the PC analysis could have been reasonably predicted by one of the two models (black; high LL). In contrast, cells left unclassified by our model-based approach were much more poorly predicted by one of the two models (gray; lower LL); this difference between the two classification approaches was statistically significant (median LL -36 vs. -32, p < 0.001, rank-sum test, nMB = 177, nPC = 244). Example traces c7–c9 illustrate cells which are differently classified by PC and our model-based approach. Gray traces indicate single-trial calcium responses; the single-trial responses used in our analysis are indicated as black dots. Blue and red curves indicate predicted single-trial distributions under the component and pattern models, respectively, generated as part of our model-based analysis. Insets in c7–c9 indicate both the model-based and PC classification of the corresponding cell. Labels in (A–C) indicate the measures corresponding to each example cell. Trace c7 shows a cell which is classified as a “pattern” cell by PC analysis, but “component” cell by our model-based analysis. The observed single-trial responses to plaid stimuli (black dots) are more likely under the component model than the pattern model (blue vs. red curves). Trace c8 shows a cell that could not be classified under PC analysis, but was classified as a component cell by our model-based analysis. Trace c9 shows a cell that was classified as a pattern cell by PC analysis, but was left unclassified by our analysis since the single-trial responses to plaid stimuli were equally well predicted by the component and pattern models (blue vs. red curves). Scale bars: 10 s and 10% ΔF/F. ***p < 0.001.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4525018&req=5

Figure 3: Model-based classification performs better than partial correlation (PC) analysis. (A) The distribution of Z-scored partial correlations for the component (Zc) and pattern (Zp) models is plotted, along with the decision boundaries used to classify responses under the PC analysis (dashed lines; see Materials and Methods). Each cell is plotted with a color and symbol according to its classification under our model-based analysis method (see text; Materials and Methods). Many of the neurons left unclassified by the PC analysis are successfully classified by our model-based approach [see cells plotted in the “Unclassified (PC)” region]. In addition, some cells classified by the PC analysis are classified differently, or left unclassified, by our model-based approach. (B) Cells classified by our model-based approach (gray) are just as well-predicted as cells classified by PC analysis (black; median LL −32 vs. −32, p = 0.88, rank-sum test, nMB = 355, nPC = 85). (C) However, cells left unclassified by the PC analysis could have been reasonably predicted by one of the two models (black; high LL). In contrast, cells left unclassified by our model-based approach were much more poorly predicted by one of the two models (gray; lower LL); this difference between the two classification approaches was statistically significant (median LL -36 vs. -32, p < 0.001, rank-sum test, nMB = 177, nPC = 244). Example traces c7–c9 illustrate cells which are differently classified by PC and our model-based approach. Gray traces indicate single-trial calcium responses; the single-trial responses used in our analysis are indicated as black dots. Blue and red curves indicate predicted single-trial distributions under the component and pattern models, respectively, generated as part of our model-based analysis. Insets in c7–c9 indicate both the model-based and PC classification of the corresponding cell. Labels in (A–C) indicate the measures corresponding to each example cell. Trace c7 shows a cell which is classified as a “pattern” cell by PC analysis, but “component” cell by our model-based analysis. The observed single-trial responses to plaid stimuli (black dots) are more likely under the component model than the pattern model (blue vs. red curves). Trace c8 shows a cell that could not be classified under PC analysis, but was classified as a component cell by our model-based analysis. Trace c9 shows a cell that was classified as a pattern cell by PC analysis, but was left unclassified by our analysis since the single-trial responses to plaid stimuli were equally well predicted by the component and pattern models (blue vs. red curves). Scale bars: 10 s and 10% ΔF/F. ***p < 0.001.
Mentions: Our model-based analysis method offers several advantages over classification of pattern and component cells using PC analysis. Firstly, we are able to classify a greater proportion of responses into pattern and component classes (304 vs. 85 cells; Figure 3A), without loss of performance (no difference in log likelihoods; Figure 3B). Model predictions were also significantly better correlated with the recorded responses of successfully classified neurons than with responses of unclassified neurons (Supplementary Figure 1; median correlations 0.55 vs. 0.08; p < 0.01, rank-sum test). For some cells, the trial-averaged response used by PC analysis was a poor description of the cell's full response, so that our model-based approach assigned a different category than PC analysis (e.g., Figure 3 trace c7). The cells left unclassified by our model-based method had responses that were poorly explained by either of the pattern or component models (low log likelihoods; Figure 3C), or were equally well explained by both models (see Figure 3 trace c9). However, cells left unclassified under the PC analysis in general had responses that fit reasonably well into one of the pattern or component models (significantly higher log likelihoods, p < 0.001; Figure 3C; see Figure 1 trace c2 and Figure 3 trace c8).

Bottom Line: We also found a large proportion of cells that respond strongly to only one stimulus class.Our results show that a broad range of pattern integration processes already take place at the level of V1.This diversity of integration is consistent with processing of visual inputs by local sub-networks within V1 that are tuned to combinations of sensory features.

View Article: PubMed Central - PubMed

Affiliation: Laboratory of Neural Circuit Dynamics, Brain Research Institute, University of Zürich Zürich, Switzerland ; Biozentrum, University of Basel Basel, Switzerland.

ABSTRACT
Neurons in sensory areas of neocortex exhibit responses tuned to specific features of the environment. In visual cortex, information about features such as edges or textures with particular orientations must be integrated to recognize a visual scene or object. Connectivity studies in rodent cortex have revealed that neurons make specific connections within sub-networks sharing common input tuning. In principle, this sub-network architecture enables local cortical circuits to integrate sensory information. However, whether feature integration indeed occurs locally in rodent primary sensory areas has not been examined directly. We studied local integration of sensory features in primary visual cortex (V1) of the mouse by presenting drifting grating and plaid stimuli, while recording the activity of neuronal populations with two-photon calcium imaging. Using a Bayesian model-based analysis framework, we classified single-cell responses as being selective for either individual grating components or for moving plaid patterns. Rather than relying on trial-averaged responses, our model-based framework takes into account single-trial responses and can easily be extended to consider any number of arbitrary predictive models. Our analysis method was able to successfully classify significantly more responses than traditional partial correlation (PC) analysis, and provides a rigorous statistical framework to rank any number of models and reject poorly performing models. We also found a large proportion of cells that respond strongly to only one stimulus class. In addition, a quarter of selectively responding neurons had more complex responses that could not be explained by any simple integration model. Our results show that a broad range of pattern integration processes already take place at the level of V1. This diversity of integration is consistent with processing of visual inputs by local sub-networks within V1 that are tuned to combinations of sensory features.

No MeSH data available.