Extended Data Figure 13 |. Incorporating P1 neural activity improves model performance.

a, Hit rate (fraction of predicted turns accompanied by a real turn; true positive rate) and false-alarm rate (fraction of predicted turns not accompanied by a real turn; false positive rate) of the naïve model versus when input current to LC10a neurons is scaled by the functional responses of P1 neurons. b, Example of the predicted turning over a courtship trial by a “naïve-model” (as in Fig. 5a) in which input current to LC10a neurons is consistently high. c-d, Two examples of actual (left) versus predicted (middle) turning responses when the activity of P1 neurons (right) is incorporated into the model. Compare to naïve model in (g). Black lines indicate when stimulus is oscillating. Details of statistical analyses and sample sizes are given in Supplementary Table 1.