Pal Debojyoti, Sharma Deepak, Kumar Mukesh, Sandur Santosh K
a Radiation Biology & Health Sciences Division, Bio-Medical Group , Bhabha Atomic Research Centre , Trombay , Mumbai , India ;
b Solid State Physics Division, Physics Group , Bhabha Atomic Research Centre , Trombay , Mumbai , India.
Free Radic Res. 2016 Sep;50(9):1011-21. doi: 10.1080/10715762.2016.1216551. Epub 2016 Aug 22.
S-glutathionylation of proteins plays an important role in various biological processes and is known to be protective modification during oxidative stress. Since, experimental detection of S-glutathionylation is labor intensive and time consuming, bioinformatics based approach is a viable alternative. Available methods require relatively longer sequence information, which may prevent prediction if sequence information is incomplete. Here, we present a model to predict glutathionylation sites from pentapeptide sequences. It is based upon differential association of amino acids with glutathionylated and non-glutathionylated cysteines from a database of experimentally verified sequences. This data was used to calculate position dependent F-scores, which measure how a particular amino acid at a particular position may affect the likelihood of glutathionylation event. Glutathionylation-score (G-score), indicating propensity of a sequence to undergo glutathionylation, was calculated using position-dependent F-scores for each amino-acid. Cut-off values were used for prediction. Our model returned an accuracy of 58% with Matthew's correlation-coefficient (MCC) value of 0.165. On an independent dataset, our model outperformed the currently available model, in spite of needing much less sequence information. Pentapeptide motifs having high abundance among glutathionylated proteins were identified. A list of potential glutathionylation hotspot sequences were obtained by assigning G-scores and subsequent Protein-BLAST analysis revealed a total of 254 putative glutathionable proteins, a number of which were already known to be glutathionylated. Our model predicted glutathionylation sites in 93.93% of experimentally verified glutathionylated proteins. Outcome of this study may assist in discovering novel glutathionylation sites and finding candidate proteins for glutathionylation.
蛋白质的S-谷胱甘肽化在各种生物过程中发挥着重要作用，并且已知在氧化应激期间是一种保护性修饰。由于S-谷胱甘肽化的实验检测劳动强度大且耗时，基于生物信息学的方法是一种可行的替代方案。现有方法需要相对较长的序列信息，如果序列信息不完整，可能会妨碍预测。在这里，我们提出了一个从五肽序列预测谷胱甘肽化位点的模型。它基于来自实验验证序列数据库中氨基酸与谷胱甘肽化和非谷胱甘肽化半胱氨酸的差异关联。该数据用于计算位置依赖性F分数，该分数衡量特定位置的特定氨基酸如何影响谷胱甘肽化事件的可能性。使用每个氨基酸的位置依赖性F分数计算谷胱甘肽化分数（G分数），以表明序列进行谷胱甘肽化的倾向。使用截止值进行预测。我们的模型返回的准确率为58%，马修斯相关系数（MCC）值为0.165。在一个独立的数据集中，尽管需要的序列信息少得多，但我们的模型优于当前可用的模型。在谷胱甘肽化蛋白质中鉴定出丰度高的五肽基序。通过指定G分数获得了潜在的谷胱甘肽化热点序列列表，随后的蛋白质BLAST分析总共揭示了254个推定的可谷胱甘肽化蛋白质，其中一些已知已被谷胱甘肽化。我们的模型在93.93%的实验验证的谷胱甘肽化蛋白质中预测了谷胱甘肽化位点。这项研究的结果可能有助于发现新的谷胱甘肽化位点并找到谷胱甘肽化的候选蛋白质。
