Training and Testing Sets of Membrane Proteins

As the set of amino acid attributes, any hydrophobicity scale or any statistical, physical, chemical, or mathematical property of the 20 natural amino acid types may be used. Many amino acid scales either are obtained experimentally (by determining properties of individual amino acids) or are computed by statistical analysis on a certain set of membrane proteins. However, it is not clear in advance which one among 

The subjective choice of attributes can be eliminated also by the procedure of testing all published amino acid scales with respect to the prediction accuracy for transmembrane helices in the chosen training data set of proteins. Our pref suite of algorithms can indeed perform such a task. Such results are beyond the scope of this chapter and will be published elsewhere. 

The set of proteins used in this work contains two lists the training list with 95 proteins (list-train), 43,336 residues (Table 1), and the testing list with 71 proteins (list-test), 28,487 residues (Table 2). 

Observed Predicted Observed Predicted Observed Predicted  

---