Sanofi-Aventis Deutschland GmbH, Frankfurt, Germany and.
Department of Proteomics and Signal Transduction, Max-Planck-Institute for Biochemistry, Martinsried, Germany.
Bioinformatics. 2016 Feb 1;32(3):424-31. doi: 10.1093/bioinformatics/btv699. Epub 2015 Dec 1.
Phosphoproteomics measurements are widely applied in cellular biology to detect changes in signalling dynamics. However, due to the inherent complexity of phosphorylation patterns and the lack of knowledge on how phosphorylations are related to functions, it is often not possible to directly deduce protein activities from those measurements. Here, we present a heuristic machine learning algorithm that infers the activities of kinases from Phosphoproteomics data using kinase-target information from the PhosphoSitePlus database. By comparing the estimated kinase activity profiles to the measured phosphosite profiles, it is furthermore possible to derive the kinases that are most likely to phosphorylate the respective phosphosite.
We apply our approach to published datasets of the human cell cycle generated from HeLaS3 cells, and insulin signalling dynamics in mouse hepatocytes. In the first case, we estimate the activities of 118 at six cell cycle stages and derive 94 new kinase-phosphosite links that can be validated through either database or motif information. In the second case, the activities of 143 kinases at eight time points are estimated and 49 new kinase-target links are derived.
The algorithm is implemented in Matlab and be downloaded from github. It makes use of the Optimization and Statistics toolboxes. https://github.com/marcel-mischnik/IKAP.git.
Supplementary data are available at Bioinformatics online.
磷酸化蛋白质组学测量广泛应用于细胞生物学，以检测信号转导动力学的变化。然而，由于磷酸化模式的固有复杂性以及缺乏关于磷酸化与功能之间关系的知识，通常不可能直接从这些测量结果推断蛋白质的活性。在这里，我们提出了一种启发式机器学习算法，该算法使用 PhosphoSitePlus 数据库中的激酶-靶标信息，从磷酸化蛋白质组学数据中推断激酶的活性。通过将估计的激酶活性谱与测量的磷酸化位点谱进行比较，还可以推导出最有可能磷酸化各自磷酸化位点的激酶。
我们将我们的方法应用于从 HeLaS3 细胞生成的人类细胞周期的已发表数据集和小鼠肝细胞中的胰岛素信号动力学。在第一种情况下，我们估计了 118 个在六个细胞周期阶段的激酶活性，并得出了 94 个新的激酶-磷酸化位点链接，这些链接可以通过数据库或基序信息进行验证。在第二种情况下，估计了 143 个激酶在八个时间点的活性，并得出了 49 个新的激酶-靶标链接。
该算法是用 Matlab 实现的，可以从 github 下载。它利用了 Optimization 和 Statistics 工具箱。https://github.com/marcel-mischnik/IKAP.git。
补充数据可在生物信息学在线获得。
