Skip to main content

Archived Comments for: pkaPS: prediction of protein kinase A phosphorylation sites with the simplified kinase-substrate binding model

Back to article

  1. Is pkaPS the best predictor for PKA phosphorylation sites prediction?

    Yu Xue, Lab of Cell Dynamics, USTC, P.R.China

    18 June 2007

    When I browse the PubMed for literature of phosphorylation sites prediction, I find this interesting article. Articles from Dr. Frank Eisenhaber's lab are always interesting and insighful. And the analysis in the article is comprehensive without tedious.

    As a major author of GPS (Group-based phosphorylation scoring method), I am thankful the authors appreciated our and others (e.g., NETPHOSK, SCANSITE and etc.) work justly and impartially. However, I should point out something for futher arguement.

    In Table 3 of pkaPS article, the values of Sn & Sp of SCANSITE, GPS and PREDPHOSPHO are taken from the published articles (Ref 14-16) rather than re-calculation. Acturally, the positive & negative control data sets are different at least in GPS and PREDPHOSPHO. For example, in Ref15(GPS article), the Sn & Sp of PREDPHOSPHO are re-calculated with the same data set for GPS, and the values are 93.60% & 91.34%. But in Kim 2004 article (Ref 16), the Sn & Sp are 88.3% & 91.1%. I have read the pkaPS article. The positive & negative data sets are different with GPS & PREDPHOSPHO. Thus, the comparison may be a little bit biased. Thus, if pkaPS is claimed to be the best predictor for PKA phosphorylation sites prediction, all algorithm might be employed and applied to the same data sets.

    Again, the article is still interesting and important. For example, the peptide used in the article is much longer than previous (-18,+23), about 42aa. This longer peptide provided more information for PKA phosphorylation sites prediction.

    In my experience, it's hard to say which predictor is the best. During the data accumlated and algorithms improved, I believe the authors of pkaPS, we and other colleages focused on computational PTMs will propel the PTMs sites prediction into a new phase.

    Competing interests

    None declared