Extended Data Fig. 3: Neuronal-space VEIs evoked strong and selective in vivo responses in target neurons while exhibiting population-decodable differences.
From: Functional bipartite invariance in mouse primary visual cortex receptive fields
a, Examples of MEI and VEIs synthesized with image diversity evaluated by Euclidean distance in pixel space (‘pixel-space’) or by cosine distance in the in silico population neuronal response space (‘neuronal-space’) for four example neurons. b, The confusion matrices showed the responses of each neuron to MEI (left) and neuronal-space VEIs (right) of 44 neurons. The responses of each neuron were normalized, and each row was scaled so the maximum response across all images equals 1. Neurons responded more strongly to their own MEI and neuronal-space VEIs (along the diagonal) compared to MEIs and neuronal-space VEIs from other neurons, respectively (two-sided permutation test, P < 10−4 for both cases). c, Neuronal-space VEIs stimulated in vivo closely to the level predicted in silico with respect to MEI (76 ± 3% versus 85%) (two-sided Wilcoxon signed-rank test, W = 8464, P = 7.7 × 10−3), with only 11.6% of all neurons showing different responses between neuronal-space VEIs and 85% of MEI (0.48% after BH correction; P < 0.05, two-sided Welch’s t-test with 32.61 average d.f.). d,e, Neuronal-space VEIs activated their target neurons more strongly than synthesized (d) and natural image (e) controls (two-sided Wilcoxon signed-rank test, W = 3441, P = 2.1 × 10−17 and W = 3466, P = 2.7 × 10−17 respectively) with 13.5% and 19.8% of all neurons showing higher responses to neuronal-space VEIs (0.0% and 2.5% after BH correction; P < 0.05, two-sided Welch’s t-test with 31.0 and 30.6 average d.f., respectively). c-e, Data were pooled over 207 neurons from five mice. f, In vivo population responses in mouse V1 L2/3 discriminated between the most dissimilar pair of neuronal-space VEIs. VEI identity in individual trials was decoded using a logistic regression classifier, with decoding accuracies across neurons (median 93%) exceeded chance level (50%, dashed line; one-sample t-test, t = 138.5, P = 1.7 × 10−180). Data were pooled over 235 neurons from three mice.
