计算机也会张冠李戴：暨南大学新文章提出蛋白质组学的新质量标准

　　

   2015年7月23日，蛋白质组学权威期刊Journal of Proteome Research在线发表了暨南大学生命与健康工程研究院何庆瑜、张弓、王通教授团队的最新研究成果，使用表观遗传干预的方法对几株人细胞进行干预，强制开放一些基因的表达，测定了32个以往从未有过蛋白质证据的“漏检蛋白”(missing protein)，使得人类向绘制完整的人类蛋白质组图谱又进了一步。在这项研究中，研究者意外发现，需要对蛋白质质谱谱图进行严格的人工质控，并谨慎选择质谱搜库软件，否则容易出现不可靠的蛋白质鉴定。

   众所周知，目前所广泛采用的鸟枪法蛋白质质谱，是鉴定复杂蛋白质样品的唯一有效方法。蛋白质样品经过酶解成为肽段，在质谱仪中被电离，生成质谱谱图，再由搜库软件对比肽段理论谱图，从而鉴定出肽段，再根据肽段信息鉴定蛋白质。由于实验存在误差、仪器也会有噪声，谱图质量和谱图的比对质量好坏，就成为鉴定可靠性的重要指标。然而在大多数研究中，人们过于相信成熟的搜库软件所给出的鉴定结果，却很少去人工检视这些被鉴定到的谱图是否真的可靠。

   在本研究中，研究者们分别使用两种被广泛使用的搜库软件Mascot和MaxQuant来处理所得到的蛋白质质谱谱图数据，意外发现，在人类蛋白质组计划所规定的FDR<1%标准下，Mascot和MaxQuant鉴定出的“漏检蛋白”中竟然分别有45%和81%无法通过人工谱图质量检查。这些低质量的谱图匹配都存在着明显的主峰不匹配、鉴定峰不连续、信噪比低等问题，仅仅由于算法的局限性才成为“漏网之鱼”。

   低得惊人的真实鉴定率促使研究者们更加仔细地对谱图进行检查，结果发现了一个真实的实例，一个本应属于Alpha-NAC muscle-specific form蛋白质的肽段由于同时具备了一个单氨基酸替换和一个化学修饰，被搜库算法错误地鉴定为Alpha-NAC pseudogene 1蛋白质。研究者通过人工检视、Smith-Waterman算法对所有肽段进行容错匹配、pLabel算法分析等手段才纠正了这一“张冠李戴”的鉴定。

   基于这些发现，为了力保鉴定质量，研究者对所有的谱图鉴定一律采用了上述人工加计算的方法进行检查，排除低质量谱图匹配、单氨基酸替换、化学修饰等因素导致的不可靠鉴定，最终得到了32个符合如此严格标准的“漏检蛋白”。

   该研究为今后的蛋白质组工作提出了新的质量标准，指出现有搜库算法由于其局限性会得到许多假阳性的鉴定结果，需要进行更严格的质控。不加分别地合并多种算法的鉴定结果将极为危险，最终会导致虚假的高鉴定量——这正是去年Nature杂志上两篇“人类蛋白质组草图”论文饱受质疑的原因。

原文摘要：

Finding missing proteins from the epigenetically manipulated human cell with stringent quality criteria

The chromosome-centric human proteome project (C-HPP) has made great progress of finding protein evidence (PE) for missing proteins (PE2-4 proteins defined by the neXtProt), which now becomes an increasingly challenging field. As a majority of samples tested in this field was from adult tissues/cells, the developmental stage specific or relevant proteins could be missed, due to biological source availability. We posit that epigenetic interventions may help to partially bypass such a limitation by stimulating the expression of the "silenced" genes in adult cells, leading to the increased chance of finding missing proteins. In this study, we established in vitro human cell models to modify the histone acetylation, demethylation and methylation with near physiological conditions. With mRNA-seq analysis, we found that histone modifications resulted in overall increases of expressed genes in an even distribution manner across different chromosomes. We identified 64 PE2-4 and 6 PE5 proteins by MaxQuant (FDR<1% at both protein and peptide levels) and 44 PE2-4 and 7 PE5 proteins by Mascot (FDR<1% at peptide level) searches, respectively. However, only 24 PE2-4 and 3 PE5 proteins in Mascot, and 12 PE2-4 and 1 PE5 proteins in MaxQuant searches could respectively pass our stringently manual spectrum inspections. Collectively, 27 PE2-4 and 5 PE5 proteins were identified from the epigenetically modified cells; among them, 19 PE2-4 and 3 PE5 proteins passed FDR<1% at both peptide and protein levels. Gene ontology analyses revealed that the PE2-4 proteins were significantly involved in development and spermatogenesis, although their chemical-physical features had no statistical difference from the background. In addition, we presented an example of suspicious PE5 peptide spectrum matched with unusual AA substitutions related to post-translational modification. In conclusion, the epigenetically manipulated cell models should be a useful tool for finding missing proteins in C-HPP. The mass spectrometry data have been deposited to the iProx database (accession number: IPX00020200).