Extended Data Fig. 1: Quantitative analysis of pattern separation in neuronal networks.
a, b, Schematic illustration of pattern separation. (a) Neuronal activity at the input (top) and the output level (bottom) during two similar contexts (top). Red, cells active in pattern A; green, cells active in pattern B; yellow, cells active in both patterns. (b) Overlay of neuronal activity at the input (top) and the output level (bottom). Highly overlapping input patterns (A, B; top) are converted into weakly overlapping output patterns (A′, B′; bottom). Modified from Johnston et al., 2016 (ref. 65). c, d, Analysis of pattern separation and pattern completion in input-output correlation plots (Rout–Rin graphs). Rin and Rout represent pairwise correlations in input and output patterns. Red dashed line indicates pattern identity. Area below identity line (red and green stripes, c) represents a regime in which Rout < Rin, that is, pattern separation. Area above identity line (yellow area, d) corresponds to a regime where Rout > Rin, that is, pattern completion. Insets, Venn diagrams of two patterns before and after pattern separation (c) and pattern completion (d). e, f, Quantitative analysis of Rout–Rin graphs. Data points (black points) represent output and input correlations for all pairs of patterns; 4950 data points total. An integral-based metric, ψ, provides a robust assessment of the average pattern separation behavior (e, main panel). ψ was computed as the area between identity line (IL, red dashed line) and the interpolated Rout–Rin curve (light gray area), normalized to the maximum area (0.5). A slope-based measure, γ, provides a selective analysis of pattern separation in a region of interest in which differences between input patterns are small (e, inset). γ was computed as the slope of the Rout–Rin curve for Rin → 1. A rank correlation-based measure, ρ, provides an analysis of the ability of the network to preserve rank order similarity (f). ρ was computed as the Pearson’s correlation coefficient of the ranks of all Rout versus the ranks of all Rin data points. Rout–Rin plot and rank correlation plots are shown for standard model parameters (same data as in Fig. 1c, f; see Supplementary Table 1).
