Extended Data Fig. 6: Further quantification of T4b PD distribution.

Related to Fig. 4. a, Fig. 4c replotted using Mercator projection. b, Visual angles subtended by T4b PD vectors (i.e., angular size). Scatter plots show the reconstructed T4b PDs (black dots, also in Fig. 4b and interpolated ones (blue dots, also in Fig. 4c) along the equator (+/−15° horizontal shaded band) and the central meridian (+/−15° vertical shaded crescent). Most T4b PDs span between 10° and 15°, but there are almost twofold differences found across the eye, with larger spans towards the rear and smaller spans near the front. c, We reconstructed all T4 types (16 ~ 20 cells) in an early pilot study at each of these four locations. d, We first mapped these T4 neurons’ PD vectors to the regular grid in Fig. 2h. Then, we computed the angles between all T4a vs T4b pairs at each location, represented here as median +/- quartiles. Similarly, for T4c vs T4d. e, Global search for optimal optic-flow fields yielded these error maps, showing the average angular differences between the T4b PD field and the optic-flow field induced by a rotation (left) or translation (right) along that direction (see Methods: ‘Ideal optic-flow fields’). Symbols “+” and “X” denote the axes of translational and rotational motion with minimal angular difference, respectively. Symbols ⨁ and ⨂ denote those with maximal differences (minimum and maximum are antipodal). f, Having established a 1-to-1 matching between two point sets in Fig. 4a, we use non-parametric regression to map new points from one space to the other. As a test of this method, we computed the pairwise distance from the regressed position of a medulla column to its matched ommatidia direction. ‘Perfect’ regression would yield zero pairwise distances, but perfect regression is impossible because the deformation is not affine. Nevertheless, these pairwise distances are a small fraction (nearly all <10%) of the inter-ommatidial angles.