Figure 4

Lamb experiment: comparison of the three transfer entropy estimation techniques. Results from applying the fixed-bin, kernel density estimation, and adaptive partitioning techniques are presented in the first, second, and third columns, respectively. Within each panel, transfer of entropy during baseline (Control) and after drug administration (Domperidone) are compared. Each closed circle represents one animal, and the box-plots capture the group statistics (median and upper and lower quartiles). The control and post-domperidone transfer entropies from the same lamb are connected with a straight line; we expected to see as many lines with a positive slope as possible. The numbers above each box-plot indicate the total number of animals who passed the surrogate test of significance. † and †† indicate p < 0.05 and p < 0.01, respectively. Note that all three techniques indicate a statistically significant increase in PO ₂ chemosensitivity following domperidone administration; fixed-bin method in panel B: 0.55 bits [0.48 0.65] ([interquartile-range]) to 0.74 bits [0.58 0.83] (p < 0.05), kernel density estimation in panel D: 0.25 bits [0.20 0.35] to 0.40 bits [0.21 0.54] (p < 0.05), and adaptive partitioning in panel F: 0.37 bits [0.30 0.41] to 0.54 bits [0.40 0.73] (p < 0.01). Note that the partitioning technique indicated the largest increase in transfer entropy post-domperidone. Furthermore, only the partitioning technique revealed a significant increase in PCO ₂ chemosensitivity post-domperidone; panel E: 0.29 bits [0.24 0.32] to 0.36 bits [0.25 0.49] (p < 0.05).