nction constructed for the factor Xa protease cleavage data [Yang,
06], where the peptides were encoded using the binary encoding
. Figure 3.33(b) and Figure 3.33(c) show the ROC curves of two
dels of the sigmoid and polynomial kernels constructed for the
ncoded factor Xa data. They show similar performance, but the
del using the polynomial kernel seems the best.
r than using the binary encoding approach for the amino acids, a
matrix can be used for transforming an amino acid space to a
ace. Having understood the powerfulness of the support vector
the bio-basis function was used as a kernel function for the
vector machine, which was named as the bio-support vector
(bio-SVM) [Yang and Chou, 2004b]. Figure 3.34 shows the ROC
the bio-SVM model constructed for the factor Xa protease
data, in which the Dayhoff mutation matrix [Dayhoff and
, 1978] was used to generate the bio-support vectors.
The ROC curve of the Dayhoff kernel bio-SVM for the factor Xa protease
ata. The AUC was 0.952.
e relevance vector machine algorithm
vance vector machine (RVM) algorithm [Tipping, 2001]
y employs the Gaussian kernel function, where ߚ is a smoothing
r for the Gaussian function, y is a vector of data labels, w is a