Computational Biology, Amgen Inc., Seattle, WA 98119-3105, USA.
Proteomics. 2010 Mar;10(6):1172-89. doi: 10.1002/pmic.200900074.
Despite recent advances in qualitative proteomics, the automatic identification of peptides with optimal sensitivity and accuracy remains a difficult goal. To address this deficiency, a novel algorithm, Multiple Search Engines, Normalization and Consensus is described. The method employs six search engines and a re-scoring engine to search MS/MS spectra against protein and decoy sequences. After the peptide hits from each engine are normalized to error rates estimated from the decoy hits, peptide assignments are then deduced using a minimum consensus model. These assignments are produced in a series of progressively relaxed false-discovery rates, thus enabling a comprehensive interpretation of the data set. Additionally, the estimated false-discovery rate was found to have good concordance with the observed false-positive rate calculated from known identities. Benchmarking against standard proteins data sets (ISBv1, sPRG2006) and their published analysis, demonstrated that the Multiple Search Engines, Normalization and Consensus algorithm consistently achieved significantly higher sensitivity in peptide identifications, which led to increased or more robust protein identifications in all data sets compared with prior methods. The sensitivity and the false-positive rate of peptide identification exhibit an inverse-proportional and linear relationship with the number of participating search engines.
尽管近年来在定性蛋白质组学方面取得了进展，但自动识别具有最佳灵敏度和准确性的肽仍然是一个困难的目标。为了解决这一不足，描述了一种新的算法，即多搜索引擎、归一化和共识。该方法使用六个搜索引擎和一个重新评分引擎来搜索 MS/MS 光谱对蛋白质和诱饵序列。在对每个引擎的肽命中进行归一化，以从诱饵命中估计的错误率后，然后使用最小共识模型推导出肽分配。这些分配是在一系列逐渐放宽的假发现率下生成的，从而能够全面解释数据集。此外，估计的假发现率与从已知身份计算出的观察到的假阳性率具有良好的一致性。与标准蛋白质数据集（ISBv1、sPRG2006）及其已发表的分析进行基准测试表明，多搜索引擎、归一化和共识算法在肽鉴定方面始终实现了显著更高的灵敏度，与先前的方法相比，所有数据集的蛋白质鉴定都增加或更稳健。肽鉴定的灵敏度和假阳性率与参与的搜索引擎数量呈反比和线性关系。
