Figure 3. | Demixed principal component analysis of neural population data

Open accessCopyright infoDownload PDF

Demixed principal component analysis of neural population data

Figure 3.

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
Figure 3.
Download figureOpen in new tabFigure 3. Demixed PCA applied to recordings from monkey PFC during a somatosensory working memory task (Romo et al., 1999).

(a) Cartoon of the paradigm, adapted from Romo and Salinas (2003). (b) Demixed principal components. Top row: first three condition-independent components; second row: first three stimulus components; third row: first three decision components; last row: first stimulus/decision interaction component. In each subplot, the full data are projected onto the respective dPCA decoder axis, so that there are 12 lines corresponding to 12 conditions (see legend). Thick black lines show time intervals during which the respective task parameters can be reliably extracted from single-trial activity (using pseudotrials with all recorded neurons), see Materials and methods. Note that the vertical scale differs across rows. Ordinal number of each component is shown in a circle; explained variances are shown as percentages. (c) Cumulative variance explained by PCA (black) and dPCA (red). Demixed PCA explains almost the same amount of variance as standard PCA. Dashed line shows an estimate of the fraction of 'signal variance' in the data, the remaining variance is due to noise in the PSTH estimates (see Materials and methods). (d) Variance of the individual demixed principal components. Each bar shows the proportion of total variance, and is composed out of four stacked bars of different color: gray for condition-independent variance, blue for stimulus variance, red for decision variance, and purple for variance due to stimulus-decision interactions. Each bar appears to be single-colored, which signifies nearly perfect demixing. Pie chart shows how the total signal variance is split among parameters. (e) Upper-right triangle shows dot products between all pairs of the first 15 demixed principal axes. Stars mark the pairs that are significantly and robustly non-orthogonal (see Materials and methods). Bottom-left triangle shows correlations between all pairs of the first 15 demixed principal components. Most of the correlations are close to zero.

DOI: http://dx.doi.org/10.7554/eLife.10989.005