Fig. 7: Correlations between Gabor and CSVA model features do not drive the variance captured by the CSVA model in OTC.

Out-of-sample variance explained by the Gabor model was regressed out of the estimation BOLD data, and the residuals were used to re-fit the CSVA model (named CSVA Gabor-controlled) resulting in new feature weights. PCA was performed on these new weights. To clarify whether any changes in feature weights or PC loadings and scores were primarily observed in early visual cortex (EVC) or were also seen in non-EVC occipital temporal cortex (OTC), we divided our OTC ROI into these two regions (see Supplementary Methods). We conducted a separate PCA within each. a–c (top row) show findings for non-EVC OTC voxels; d–f (bottom row) show findings for EVC voxels. a, d. A bar and whisker plot shows voxel-wise correlations between feature weights from the CSVA and CSVA-Gabor-Controlled models for each subject across non-EVC OTC voxels and EVC voxels. Box plot elements: center line=median; box limits=upper and lower quartiles; whiskers=1.5 x inter-quartile range; individual points=feature weight correlation coefficients (across features) for each voxel. b, e A correlation matrix shows the similarity between feature loadings for the top 3 group PCs from the original CSVA model and the top 3 group PCs for the CSVA Gabor-controlled model for PCA conducted across non-EVC OTC voxels and EVC voxels. c, f A bar-and-whisker plot shows the correlations between PC scores from the original CSVA model and PC scores from the Gabor-Controlled model across non-EVC OTC voxels and EVC voxels. Box plot elements: as for panels a and d except individual points = PC score correlation coefficients (across voxels) for each subject.