Figure 6. | Automatic discovery of cell types and microcircuitry from neural connectomics

Open accessCopyright infoDownload PDF

Automatic discovery of cell types and microcircuitry from neural connectomics

Figure 6.

Affiliation details

University of California, Berkeley, United States; Northwestern University, United States; Rehabilitation Institute of Chicago, United States
Figure 6.
Download figureOpen in new tabFigure 6. Visualizing type inference uncertainty.

Our fully Bayesian model gives a confidence estimate (posterior probability) that any two given cells are of the same type. In (A) we visualize that cell–cell coassignment matrix, showing the probability that cell i is of the same type as cell j on a range from 0.0 to 1.0. The block structure shows subsets of cells which are believed to all belong to the same type. For comparison, (B) shows the anatomist-defined type for each cell, grouped broadly into the coarse types identified in the previous panel. (C) Link versus cluster accuracy. (D) The posterior distribution of receiver operating characteristic (ROC) curves from 10-fold cross-validation when predicting connectivity, as well as (E) the area under the curve (AUC) and (F) the type agreements with known neuroanatomist types. ARI: adjusted Rand index. Model comparison, showing using human-discovered types with and without distance information, as well as our model incorporating just connectivity, connectivity and distance, or connectivity, distance, and synaptic depth (as well as the alternative latent position cluster model, see text). (G) A comparison of the predictive accuracy (AUC) for hand-labeled anatomical data, versus inclusion of additional sources of information, as well as the clustering accuracy. Note that our model sacrifices very little predictive accuracy for additional clustering accuracy. By comparison, conventional methods fail at one or both. ARI: adjusted Rand index. (H) The spatial extent (in depth and area) of the types identified by humans and our various algorithmic approaches.

DOI: http://dx.doi.org/10.7554/eLife.04250.007