Department of Biochemistry, Brigham Young University , 701 East University Parkway, BNSN C100, Provo, Utah 84602, United States.
J Proteome Res. 2013 Dec 6;12(12):5742-9. doi: 10.1021/pr400727e. Epub 2013 Oct 3.
The most important step in any quantitative proteomic pipeline is feature detection (aka peak picking). However, generating quality hand-annotated data sets to validate the algorithms, especially for lower abundance peaks, is nearly impossible. An alternative for creating gold standard data is to simulate it with features closely mimicking real data. We present Mspire-Simulator, a free, open-source shotgun proteomic simulator that goes beyond previous simulation attempts by generating LC-MS features with realistic m/z and intensity variance along with other noise components. It also includes machine-learned models for retention time and peak intensity prediction and a genetic algorithm to custom fit model parameters for experimental data sets. We show that these methods are applicable to data from three different mass spectrometers, including two fundamentally different types, and show visually and analytically that simulated peaks are nearly indistinguishable from actual data. Researchers can use simulated data to rigorously test quantitation software, and proteomic researchers may benefit from overlaying simulated data on actual data sets.
在任何定量蛋白质组学管道中，最重要的步骤是特征检测（又名峰提取）。然而，生成质量可手动注释的数据集来验证算法，特别是对于较低丰度的峰，几乎是不可能的。创建黄金标准数据的另一种方法是使用与真实数据非常相似的特征来模拟它。我们介绍了 Mspire-Simulator，这是一个免费的、开源的 shotgun 蛋白质组学模拟器，它通过生成具有真实 m/z 和强度变化以及其他噪声成分的 LC-MS 特征，超越了之前的模拟尝试。它还包括用于保留时间和峰强度预测的机器学习模型，以及用于根据实验数据集自定义拟合模型参数的遗传算法。我们表明，这些方法适用于来自三种不同质谱仪的数据，包括两种完全不同类型的质谱仪，并通过视觉和分析表明，模拟的峰与实际数据几乎无法区分。研究人员可以使用模拟数据来严格测试定量软件，蛋白质组学研究人员可能会受益于将模拟数据叠加在实际数据集上。
