Discussion | We were able to eXpand ITK_CYCLE to search for asymmetric waveforms without degrading sensitivity because we empirically calculate p-values, which yields much more accurate significance estimates than the Bonferroni correction employed in the original formulation of ITK_CYCLE [26]. |
E 3 A A g Time s 'r r a E A AA Time Time | employed the Bonferroni correction [34] in their original formulation and implementation of the method. |
E 3 A A g Time s 'r r a E A AA Time Time | The previous version of ITK_CYCLE corrects for underestimating the p-values with the Bonferroni correction , which controls the family-wide error rate (FWER) by multiplying the p-values by the number of hypothesis tests being performed. |
E 3 A A g Time s 'r r a E A AA Time Time | A common alternative to the Bonferroni correction is the Benjamini-Hochberg procedure [36], which seeks to control the false discovery rate (FDR). |
Microarray metadataset | As experimental sampling rates and sampling densities enable more extensive searching of phases, periods, and asymmetries, we expect the advantage for empirical ITK_CYCLE relative to the original formulation to grow because the Bonferroni correction strongly penalizes adding hypothesis tests. |
Microarray metadataset | The original ITK_CYCLE method with Bonferroni correction and empirical ITK_CYCLE method identify sets of enriched genes known to have alternative splice forms of their RNA; ITK_CYCLE with the Benjamini-Hochberg correction and empirical ITK_CYCLE identify sets of genes that are enriched for genes involved in biosynthetic pathways. |
Simulated data benchmarks | The ITK_CYCLE with Benjamini-Hochberg correction (ITK_BH) has AUROC values that are in between the AUROC values for the original ITK_CYCLE with Bonferroni correction (ITK) and empirical ITK_CYCLE. |
Supporting Information | Comparison of the p-Value distributions of the original IT K_CYCLE method (with Bonferroni correction ) with the empirical IT K_CYCLE method without (A) and with (B) asymmetry search. |
Methods). | 7a, mixed 2 29.7 spk/s, nonsync = 13.9 spk/ s, sync = 3.3 spk/s; Wilcoxon rank sum test, P< 1.2 x 10'76, Bonferroni corrected ). |
Methods). | We also observed a significant difference between non-synchronized and synchronized neurons (Wil-coxon rank sum test, P< 6.9 x 10'96, Bonferroni corrected ). |
Methods). | 7b, mixed 2 51.3 spk/s, nonsync = 22.5 spk/s, sync = 18.3 spk/s; Wilcoxon rank sum test, P< 0.003, Bonferroni corrected ). |
Learning rates | The trial at which the t-test indicated that the normalised learning extent after that trial was significantly above zero (at an a-level of 0.000625 for Bonferroni correction ) was taken as the time point at which participants realised the role of the shape and started to learn the mappings. |
inter Iation test sha es '—' extrapolation W p extrapolation test shape shape parameter p test shape | As it turns out (after Bonferroni correction ), the results for all participants but one (n = 6) conformed with this linear hypothesis. |
inter Iation test sha es '—' extrapolation W p extrapolation test shape shape parameter p test shape | It is important to note that also the Mixture-of-Kalman-Filters performs quite well in predicting the generalisation behaviour although it sometimes slightly overestimates the learning extent for the training shapes (difference between yellow line and red dots) leading to significant differences in the residuals for 3 out of 7 participants after Bonferroni correction . |