Liu Huayu, Polotsky Alex J, Grunwald Gary K, Carlson Nichole E
a Eli Lilly and Company , Indianapolis , IN , USA.
b Department of Obstetrics and Gynecology, School of Medicine , University of Colorado Anschutz Medical Campus , Aurora , CO , USA.
Syst Biol Reprod Med. 2018 Feb;64(1):80-91. doi: 10.1080/19396368.2017.1411541. Epub 2017 Dec 29.
Pulsatile secretion of hormones in the hypothalamic-pituitary-gonadal axis is critical for normal functioning of the reproductive system. Thus, appropriate characterization of pulsatile secretion is important for identifying the (patho)physiology of reproductive conditions. Existing analysis methods often fail to adequately characterize pulsatility, especially when the signal-to-noise ratio is low. Newer Bayesian analysis methods for pulsatile hormones may offer improved secretion quantification in noisier data. The objective of this study was to extensively validate a Bayesian analysis approach for analyzing pulsatile hormones in settings that occur in reproductive studies. An investigative approach was chosen so that clinical research teams will have the knowledge to adopt this newer analysis approach in practice. Three experimental conditions were investigated: luteinizing hormone (LH) profiles in ovariectomized ewes (N=6; high signal-to-noise setting), LH profiles in young ovulating women (N=12; lower signal-to-noise setting), and computer-simulated scenarios (N=200). For each experimental condition, differences in luteinizing hormone pulse outcomes (pulse number, average pulse size, hormone half-life, and non-pulse secretion) were obtained and compared between non-Bayesian and Bayesian analysis pulse analysis methods. For the ewe model, the estimated pulse number and mass were comparable between the Bayesian and non-Bayesian analyses. For the human model, only 4 of 12 subjects could be fitted with the non-Bayesian analysis compared to 10 of the 12 with Bayesian analysis. In general, the Bayesian analysis had lower false negative rates (<4.5%) compared to the non-Bayesian analysis while maintaining a high specificity (false positive rate <2.5%). The Bayesian analysis also had less biased estimates of all pulse features. In conclusion, Bayesian analysis provides a more reliable pulse characterization in low signal-to-noise experiments and should be used for the analysis of reproductive physiology studies of pulsatile hormones. Software is available at www.github.com/BayesPulse .
LH: luteinizing hormone; FSH: follicle stimulating hormone; GnRH: gonadotropin-releasing hormone; FP: false positive; FN: false negative.
下丘脑 - 垂体 - 性腺轴中激素的脉冲式分泌对于生殖系统的正常功能至关重要。因此，对脉冲式分泌进行恰当的特征描述对于识别生殖疾病的（病理）生理学特征很重要。现有的分析方法常常无法充分表征脉冲性，尤其是在信噪比很低时。针对脉冲式激素的更新的贝叶斯分析方法可能在噪声更大的数据中提供更好的分泌量量化。本研究的目的是在生殖研究中的各种情况下广泛验证一种用于分析脉冲式激素的贝叶斯分析方法。我们选择了一种研究方法，以便临床研究团队能够在实践中掌握采用这种更新的分析方法的知识。研究了三种实验条件：去卵巢母羊的促黄体生成素（LH）曲线（N = 6；高信噪比情况）、年轻排卵女性的LH曲线（N = 12；较低信噪比情况）以及计算机模拟场景（N = 200）。对于每种实验条件，获取了促黄体生成素脉冲结果（脉冲数、平均脉冲大小、激素半衰期和非脉冲分泌）的差异，并在非贝叶斯和贝叶斯分析脉冲分析方法之间进行比较。对于母羊模型，贝叶斯分析和非贝叶斯分析之间估计的脉冲数和质量相当。对于人类模型，12名受试者中只有4名能够用非贝叶斯分析拟合，而贝叶斯分析为12名中的10名。总体而言，与非贝叶斯分析相比，贝叶斯分析具有更低的假阴性率（<4.5%），同时保持高特异性（假阳性率<2.5%）。贝叶斯分析对所有脉冲特征的估计偏差也更小。总之，贝叶斯分析在低信噪比实验中提供了更可靠的脉冲特征描述，应用于脉冲式激素的生殖生理学研究分析。软件可在www.github.com/BayesPulse获取。
LH：促黄体生成素；FSH：促卵泡激素；GnRH：促性腺激素释放激素；FP：假阳性；FN：假阴性。
