Fig 3. Summary of STRF gain thresholding results.

(A) Mean ± SEM prediction accuracy at each gain threshold (p_gain) minus prediction accuracy obtained with the raw STA (MU: r = 0.176, SU: r = 0.210). (B) Histograms of gain thresholds that yielded the highest prediction accuracy. (C–D) Cumulative distribution functions of prediction accuracy obtained with raw STAs, STRFs corrected with a fixed p_gain setting (p_gain < 0.01 applied uniformly across units), and STRFs corrected at the best p_gain settings (identified on an individual-unit basis through cross-validation). Inset bar plots represent mean +SEM prediction accuracy. For both data types, accuracy for each gain thresholding approach (fixed p_gain and best p_gain) was significantly higher than the raw STA, but these approaches significantly did not differ from each other (Wilcoxon signed-rank tests). (E) Scatter plots of prediction accuracy for individual units at best p_gain versus raw (left) and fixed p_gain (right). Note that prediction accuracy may be higher for the fixed p_gain setting than best p_gain setting. This is because best p_gain settings determined by the validation dataset may not maximize prediction accuracy for the test dataset (see Methods for additional details). Red markers indicate the means. (F) Mean ±SEM prediction accuracy as a function of the temporal bin size for the PSTHs and predictions using the same correction approaches and color schemes as in (C–D).