Table 1. | Demixed principal component analysis of neural population data

Open accessCopyright infoDownload PDF

Demixed principal component analysis of neural population data

Table 1.

Affiliation details

Champalimaud Centre for the Unknown, Portugal; École Normale Supérieure, France; Centre for Integrative Neuroscience, University of Tübingen, Germany; Wake Forest University School of Medicine, United States; Cold Spring Harbor Laboratory, United States; Universidad Nacional Autónoma de México, Mexico; El Colegio Nacional, Mexico; Harvard University, United States
Table 1.

Demixed PCA in comparison with existing methods. Columns: 'Signif.' refers to the method of counting significantly tuned cells, as shown in Figure 1c–e. TDR refers to the 'targeted dimensionality reduction' of Mante et al. (2013) shown in Figure 1f–h. LDA stands for linear discriminant analysis, but this column applies to any classification method (e.g. support vector machine, ordinal logistic regression, etc.). All classification methods can be used to summarize population tuning via a time-dependent classification accuracy (e.g. Meyers et al., 2012). PCA stands for principal component analysis, as shown in Figure 1i–k. FA stands for factor analysis, GPFA for Gaussian process factor analysis (Yu et al., 2009), LDS for hidden linear dynamical system (Buesing et al., 2012a; 2012b), jPCA is the method introduced in Churchland et al. (2012) . Some of the existing methods can be extended to become more general, but here we refer to how these methods are actually used in the original research. Rows: The first two rows are the two defining goals of dPCA. Following rows highlight notable features of other methods.


Takes task parameters into account & provides summary statistics of population tuning
Allows to reconstruct neural firing (captures variance)
Based on dynamical model
Based on probabilistic model
Takes spike trains as input