Extended Data Fig. 9: Results are very similar when taking the linear parameters β₀ and β₁ across cross-validated partitions.

Same format as Fig. 3 in the main text. In the main text analyses, we re-fit the linear parameters β₀ and β₁ in the test data during cross-validation. As explained in the Methods, this is to isolate our analysis to the asymmetries in scaling, rather than the analysis being impacted by, for example, overall (non-asymmetric) gain. Here we show the same result as in Fig. 3c in the same 94 neurons, but when carrying over the linear parameters (β₀ and β₁) as well as the asymmetric parameters (S, α⁺ and α−) to predict firing rate, and therefore do not re-estimate the linear parameters in the test data. We find that the pattern of results remains the same. Error bars denote SEM. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001.