There are many hydrophobicity scales for protein analysis.

Broadly, I gather the differences between them are from the experimental method to acquire the data and the normalisation (or lack thereof) of the data.

To detect/predict transmembrane segments one method is to use a window of 19-20 of these residue scales to detect a region of hydrophobicity above a threshold. The Eisenberg et al. scale is used by uniprot for example.

Uniprot don't seem to explain their choice of using the Eisenberg et al. scale.

Why might one use a certain scale over another when predicting transmembrane segments?

  • 1
    $\begingroup$ In addition to hydrophobicity, alpha-helices formation should be considered. $\endgroup$
    – 243
    Jul 6, 2015 at 5:52

1 Answer 1


This has been studied by some of my labmates in Why is the biological hydrophobicity scale more accurate than earlier experimental hydrophobicity scales? I am not involved in their research, but here is the gist of the paper:

Different scales are, as you say, developed based on different criteria. In particular, Eisenberg scale is one of the consensus scales, that is hopefully at least as good as all of its components. It was believed that there were no big performance differences between scales. This could be why Uniprot doesn't specify their choice of scale: at the time, it didn't matter too much. A study in 2009 showed that the new UHS scale was actually better than others.

The results of the benchmark are that the differences are indeed not so big: roughly $\pm2\%$ in difference in prediction accuracy. The overall winner? Hessa, scoring $1.29\ \sigma$ over the average of the scales, but closely followed by UHS, PM1D and PM3D. Eisenberg performance, by the way, is average.

Final note: they found that a near optimal window size for all the scales was 21 with double the weight in the central third.

  • $\begingroup$ This is exactly what I was after. Thanks. I don't have access to the journal at the moment. Does the article comment on Kyte & Doolittle's, or von Hiejne's scale? $\endgroup$
    – James
    Jul 6, 2015 at 8:18
  • 1
    $\begingroup$ @GoodGravy Kyte yes, and it is similar in results to Esenberg; von Heijne no (but it is indirectly present, as it is one of the components of Eisenberg). $\endgroup$
    – Davidmh
    Jul 6, 2015 at 9:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.