From 1980 to 1991, Kyriacou, Hall, and collaborators (K&H) reported that the Drosophila melanogaster courtship song has a 1-min cycle in the length of mean interpulse intervals (IPIs) that is modulated by circadian rhythm period mutations. In 2014, Stern failed to replicate these results using a fully automated method for detecting song pulses. Manual annotation of Stern’s song records exposed a ~50% error rate in detection of IPIs, but the corrected data revealed period-dependent IPI cycles using a variety of statistical methods. In 2017, Stern et al. dismissed the sine/cosine method originally used by K&H to detect significant cycles, claiming that randomized songs showed as many significant values as real data using cosinor analysis. We first identify a simple mathematical error in Stern et al.’s cosinor implementation that invalidates their critique of the method. Stern et al. also concluded that although the manually corrected wild-type and per^L mutant songs show similar periods to those observed by K&H, each song is usually not significantly rhythmic by the Lomb-Scargle (L-S) periodogram, so any genotypic effect simply reflects “noise.” Here, we observe that L-S is extremely conservative compared with 3 other time-series analyses in assessing the significance of rhythmicity, both for conventional locomotor activity data collected in equally spaced time bins and for unequally spaced song records. Using randomization of locomotor and song data to generate confidence limits for L-S instead of the theoretically derived values, we find that L-S is now consistent with the other methods in determining significant rhythmicity in locomotor and song records and that it confirms period-dependent song cycles. We conclude that Stern and colleagues’ failure to identify song cycles stems from the limitations of automated methods in accurately reflecting song parameters, combined with the use of an overly stringent method to discriminate rhythmicity in courtship songs.

从1980年到1991年，Kyriacou，Hall和合作者（K＆H）报告说，果蝇的求爱歌曲在平均脉冲间隔（IPI）的长度上有1分钟的周期，该周期由昼夜节律周期突变调节。在2014年，Stern无法使用全自动方法检测歌曲脉冲来复制这些结果。斯特恩的歌曲记录手册中标注的IPI检测暴露了〜50％的错误率，但修正后的数据显示期多种统计方法来确定依赖IPI的周期。在2017年，Stern等人。取消了K＆H最初用于检测重要周期的正弦/余弦方法，声称使用余弦分析，随机歌曲显示的真实值与真实数据一样多。我们首先在Stern等人的余弦实现中确定一个简单的数学错误，使他们对该方法的批评无效。斯特恩等。还得出结论，尽管手动校正的野生型和每^大号变异歌曲的周期与K＆H观察到的周期相似，每首歌曲通常在Lomb-Scargle（LS）周期图上没有明显的节奏感，因此任何基因型效应都仅反映了“噪音”。在这里，我们观察到LS与其他3种时间序列分析相比在评估节奏的重要性方面非常保守，无论是对于等间隔时间仓中收集的常规运动活动数据还是对于等间隔歌曲记录而言，LS的节奏性都很重要。使用运动和歌曲数据的随机化来生成LS的置信度极限，而不是从理论上得出的值，我们发现LS现在与确定运动和歌曲记录中重要的节奏性的其他方法一致，并且可以确认周期依赖的歌曲周期。我们得出的结论是，斯特恩及其同事未能确定歌曲周期是由于自动方法无法准确反映歌曲参数，以及使用过于严格的方法来区分求爱歌曲的节奏性所致。