fMRI adaptation pattern similarity analysis

The primary goal of this study was to investigate whether or not the neural representations of the four test locations were specific to or independent of the underlying cue type (self-motion vs landmark). Since the voxel-wise adaptation effects served as an index of spatial coding, we investigated cue specificity of this coding with an adaptation pattern similarity analysis. This analysis is analogous to multivariate representational similarity analysis, but with voxel-wise adaptation magnitude instead of voxel-wise activation level as the basic element. If the voxel-to-voxel adaptation pattern differed between landmarks and self-motion cues, this would suggest dissociable neural representations evoked by the two cue types.

The adaptation pattern similarity analysis consisted of several steps. First, an adaptation vector was estimated for each of the 16 runs for each cue type, containing adaptation estimates (signed) of all voxels in a given ROI. This resulted in a total of eight adaptation vectors per cue type. The secondary factor “environment” was averaged out by calculating the mean of two adaptation vectors from adjacent runs from different environments for the same cue type on a voxel-by-voxel basis (see a similar treatment of averaging out a factor not of primary interest in Shine et al., 2019).

Next, the adaptation pattern similarity was computed in a cross-validated manner by calculating Pearson’s correlation between pairwise adaptation vectors (Walther et al., 2016). Specifically, within-cue similarity was calculated as the Pearson’s correlation between adaptation vectors of the same cue type. In contrast, between-cue similarity was computed by correlating adaptation vectors from different cue types. The final estimates of within-cue similarity and between-cue similarity were obtained by averaging all Pearson’s correlations (Fisher transformed) calculated from all possible pairs of adaptation vectors.

Finally, we calculated an adaptation pattern distinction score by subtracting the between-cue similarity from the within-cue similarity. The adaptation pattern distinction score was then tested against zero with a directional one-sample t test, as we expected the adaptation pattern to be more similar within the same cue type than between different cue types. A positive adaptation pattern distinction score indicates that the spatially distributed adaptation pattern differed between different cue types.

To verify that our findings were not solely driven by a particular environment, we also conducted the analysis for each environment separately.