Fig. 6: MICrONS functional connectomics analysis.
From: Functional bipartite invariance in mouse primary visual cortex receptive fields
a, Schematic of the MICrONS functional connectomics dataset8, comprising responses of >75,000 neurons to dynamic stimuli and their reconstructed subcellular connectivity from electron microscopy data. We employed the MICrONS ‘digital twin’9, trained on dynamic stimuli (denoted as a ‘dynamic’ model; recurrent neural network, RNN) to predict responses to natural images used in our experiments. A new CNN model was trained on these in silico predictions (‘dynamic-static’ or DS model) and used to synthesize MEIs, VEIs and VEIspartial. b, MEIs and VEIs optimized using our standard model (‘static’ or S model) and DS model for two example neurons. c, DS-VEIs stimulated neurons in vivo at 80 ± 3% of DS-MEI activation, close to the in silico prediction of 85% (two-sided Wilcoxon signed-rank test, W = 31,534, P = 2.8 × 10−4), with only 10.3% of all neurons showing different responses between VEIs and 85% of MEI (0.25% after BH correction) (P < 0.05, two-sided Welch’s t-test with 32.0 average d.f.). d, DS-VEIspartial activated target neurons similarly to DS-VEIs (two-sided Wilcoxon signed-rank test, W = 29,878, P = 1.4 × 10−5) with only 9.5% of all neurons showing different responses (0.0% after BH correction) (P < 0.05, two-sided Welch’s t-test with 32.0 average d.f.). e, DS-MEIs were more similar to S-MEIs of the same neuron than S-MEIs of other random neurons (two-sided Wilcoxon signed-rank test, W = 4,537, P = 4.0 × 10−53). f, Similarly, DS-VEIs were more similar to S-VEIs of the same neuron than S-VEIs of other random neurons (two-sided Wilcoxon signed-rank test, W = 3,969, P = 8.8 × 10−55). g, The mean MEI and VEI similarities of connected pairs (0.06 ± 0.02 and 0.04 ± 0.02) were higher than those of the ADP control pairs26 (0.03 ± 0.01 and 0.021 ± 0.004; P = 0.02 and P < 10−4, respectively, two-sided bootstrapped mean difference after BH correction). h,i, Synapse conversion rate (Nsyn/mm Ld where Nsyn denotes the number of synapses between two neurons and Ld denotes the axon-dendrite co-travel distance in mm) increased linearly with the MEI (h) and VEI (i) representational similarity for neuron pairs (P = 0.014 and 0.0034, respectively, two-sided t-test for linear coefficient against 0 using Poisson generalized linear mixed model with random intercepts). Neuron pairs were binned by their MEI and VEI similarity, respectively. Shaded areas represented 95% CIs from bootstrapping. j, Diversity indices from the DS model highly correlated with those from the S model (Pearson r = 0.46, P = 1.2 × 10−22, two-sided t-test). k, The mean diversity index increase for connected pairs was greater than that for ADP control pairs (0.16 ± 0.02 and 0.14 ± 0.01, respectively; P = 0.04, two-sided bootstrapped mean difference against 0 after BH correction). l, Presynaptic neurons with lower diversity indices showed higher synapse conversion rate (Spearman’s rank correlation coefficient ρ = − 0.49, P = 0.03, two-sided t-test). This relationship was well-modeled by an exponential decay (R2 = 0.58). g,k, Box plots show center line (median); box bounds (25th to 75th percentiles, IQR); whiskers extend to the most extreme data points within 1.5 × IQR of the quartiles; caps mark whisker ends; points beyond whiskers are plotted as outliers. c–f,j, Data for in vivo verification of the DS model were pooled over 399 neurons from three mice. g–i,k,l, Data for MICrONS functional connectomics analysis were pooled over 19 presynaptic neurons forming 706 connected pairs and 18,162 ADP controls.
