Fig. 2: Donut-like inhibition surpasses other hypothesized circuit motifs for generating categorical responses in models of the avian midbrain selection network.

A Computational models incorporating combinations of three different circuit motifs proposed in the literature as underlying categorical responses (Methods; Fig. S1). All models built upon generic “baseline” circuit (left column-top) capable of comparing competing options. Each model shows two “channels”; each channel is the group of neurons (numbered) involved in representing a stimulus (S1 or S2). Large circles: OTid neurons; ovals: Imc neurons; small circles: Ipc neurons. Arrows with pointed heads - excitatory connections, arrows with flat heads - inhibitory connections. The three circuit motifs are – feedback inhibition between competing channels (green; middle column - top), donut-like inhibition (self-sparing inhibition; purple; middle column-second from top; see also text), and recurrent amplification within each channel (through “Ipc” neuron; orange; left column-bottom). The goal of output neuron 1 (bold face) in each model is to signal if S1 > S2 (category “a”) or S2 > S1 (category “b”), when presented with S1 and S2 of varying relative strength following strength-morphing protocol. B Strength-morphing protocol (Methods). S1 and S2 are presented simultaneously “to” the model, S1(2) is inside the receptive field of neurons in channel 1 (2). As strength of S1 is decreased, that of S2 is systematically increased; strength of the stimuli is controlled by their loom speeds (°/s); denoted here by size of dots. Gray line: ideal selection boundary (when relative strength = 0). C Simulated response profiles of output neuron 1 from each of the models (colors) in A obtained using the strength- morphing protocol (bottom inset). Responses are mean ± s.e.m of 30 repetitions. The continuously varied input parameter was the relative strength of the two stimuli (S2-S1). Lines – best sigmoidal fits. Input-output functions of model neurons were sigmoids with Gaussian noise (Methods; fano factor=6). Right-Inset: Response profiles normalized between 0 and 1; only means are shown for clarity. D “Population” summary of CatI of response profiles from various circuit models (colors); n = 50 model neurons; center lines in the violin plots indicate median values. *p < 0.05, One-way ANOVA followed by Holm-Bonferroni correction for multiple paired comparisons; only a key subset of significant differences indicated for clarity. green vs. gray: p = 0.99, purple vs. gray: p = 5.98e-8, blue vs. purple: p = 5.98e-8, red vs. purple. See also Fig. S1. Source data are provided as a Source Data file.