In absence of periodic systematic comparisons, biologists/bioinformaticians may be forced to make a subjective selection among the many protein-protein interaction (PPI) databases and tools. We conducted a comprehensive compilation and comparison of such resources. We compiled 375 PPI resources, short-listed 125 important ones (both lists are available at startbioinfo.com), and compared the features and coverage of 16 carefully-selected databases related to human PPIs. We quantitatively compared the coverage of 'experimentally verified' as well as 'total' (experimentally verified and predicted) PPIs for these 16 databases. Coverage was compared in two ways: (a) PPIs obtained in response to gene queries using the web interfaces were compared. As a query set, 108 genes expressed differently across tissues (specific to kidney, testis, and uterus, and ubiquitous - i.e., expressed in 43 human normal tissues) or associated with certain diseases (breast cancer, lung cancer, Alzheimer's, cystic fibrosis, diabetes, and cardiomyopathy) were chosen. The coverage was also compared for the well-studied genes versus the less-studied ones. The coverage of the databases for high-quality interactions was separately assessed using a set of literature curated experimentally-proven PPIs (gold standard PPI-set); (b) the back-end-data from 15 PPI databases was downloaded and compared. Combined results from STRING and UniHI covered around 84% of 'experimentally verified' PPIs. Approximately 94% of the 'total' PPIs available across the databases were retrieved by the combined use of hPRINT, STRING, and IID. Among the experimentally verified PPIs found exclusively in each database, STRING contributed around 71% of the hits. The coverage of certain databases was skewed for some gene-types. Analysis with the gold-standard PPI-set revealed that GPS-Prot, STRING, APID, and HIPPIE, each covered ~70% of the curated interactions. The database usage frequencies did not always correlate with their respective advantages, thereby justifying the need for more frequent studies of this nature.

中文翻译：

---

从用户角度对蛋白质-蛋白质相互作用数据库进行系统比较。

在没有定期系统比较的情况下，生物学家/生物信息学家可能被迫在许多蛋白质-蛋白质相互作用（PPI）数据库和工具中进行主观选择。我们对这些资源进行了全面的汇编和比较。我们收集了375个PPI资源，入围了125个重要PPI资源（这两个列表都可以在startbioinfo.com上找到），并比较了16种与人类PPI相关的精心选择的数据库的功能和覆盖范围。我们定量比较了这16个数据库的“经过实验验证”以及“总计”（经实验验证和预测）的PPI的覆盖范围。通过两种方式比较了覆盖率：（a）比较了使用网络界面响应基因查询而获得的PPI。作为一个查询集，有108个基因在不同的组织中表达不同（特定于肾脏，睾丸，选择子宫，子宫以及无处不在（即在43个人类正常组织中表达）或与某些疾病（乳腺癌，肺癌，阿尔茨海默氏病，囊性纤维化，糖尿病和心肌病）相关的疾病。还对研究好的基因与研究较少的基因的覆盖率进行了比较。使用一组精选的经过实验验证的PPI（黄金标准PPI集），分别评估了高质量交互作用的数据库覆盖范围；（b）从15个PPI数据库中下载了后端数据并进行了比较。STRING和UniHI的合并结果涵盖了大约84％的“经过实验验证”的PPI。通过组合使用hPRINT，STRING和IID，可检索整个数据库中可用的“全部” PPI的约94％。在每个数据库中唯一找到的经过实验验证的PPI中，STRING贡献了约71％的匹配。对于某些基因类型，某些数据库的覆盖范围存在偏差。使用黄金标准PPI集进行的分析表明，GPS-Prot，STRING，APID和HIPPIE各自覆盖了约70％的策划互动。数据库使用频率并不总是与它们各自的优势相关，从而证明需要更频繁地研究这种性质。