Fig. 3.

Prediction of the connectivity with the NNC and emergence of LN groups. (A) Correlation between the four ORNs $\to$ LN connection weight vectors $\{{\mathbf{w}}_k\}$ from NNC-4 ($\rho =1$) and the four ORNs $\to$ LN_type synaptic count vectors $\left\{w_{\text{LNtype}}\right\}$ (SI Appendix, Fig. S6 C, D, F, G, and H). One-sided P-values calculated by shuffling the entries of each ${\mathbf{w}}_{\text{LNtype}}$ (50,000 permutations). *: significant at 0.05 FDR. (B) Bottom: maximum correlation coefficient (mean $\pm$ SD) of the four ${\mathbf{w}}_k$s from NNC-4 with the four ${\mathbf{w}}_{\text{LNtype}}$s for different values of $\rho$ (50 simulations per $\rho$), encoding the feedback inhibition strength. Top: number of ${\mathbf{w}}_{\text{LNtype}}$s significantly correlated with at last one ${\mathbf{w}}_k$ from NNC-4 (FDR at 5%). For $\rho ⪆3.1$, not all simulations converge to the same $\{{\mathbf{y}}^{(t)}\}$, $\{{\mathbf{z}}^{(t)}\}$, and $\{{\mathbf{w}}_k\}$, potentially due to existence of multiple global optima or simulations only finding local optima. (C) Correlation between the $w_{\text{LN}}$s on the left and right sides of the larva, portraying that several $w_{\text{LN}}$s are similar. (D) Same as (C) for the eight ${\mathbf{w}}_k$s arising from NNC-8 and with $\rho =0.1,0.35,1,10$. Matrices ordered using hierarchical clustering and ${\mathbf{w}}_k$s ordered accordingly (SI Appendix). (E) Mean rectified correlation coefficient ${\overline{r}}_{+}$ ($r_{+}:=max[0,r]$) from (C) (blue band delimited by the value for left and right circuit) and from NNC-8 (black line, mean $\pm$ SD, 50 simulations per $\rho$). ${\overline{r}}_{+}$ obtained by averaging all the $r_{+}$ from a correlation matrix, i.e., (C) or (D), excluding the diagonal.