| Method |
Parameters |
Self-Prediction |
Cross-Validation |
| MARS |
Degree=1, Penalty=1: Used:
GBACC(1), INTERPRO(1), KEGG(1), MEDLINE(2), MIM(1), SPID(1)
GCI = 5.417 + 0.5127 * max(X1 - 1.091, 0)
- 1.328 * max(1.091 - X1, 0) + 0.4746 * max(X2 + 0.723, 0)
+ 0.3539 * max(X3 + 0.2628, 0) + 0.2082 * max(X4 + 0.7626, 0)
+ 0.2478 * max(X5 + 0.6658, 0) - 0.395 * max(1.508 - X6, 0)
where: X1 = ( log(1+MEDLINE) - 1.322 ) / 1.139
X2 = ( min(SPID,2) - 0.3532 ) / 0.4885
X3 = ( min(KEGG,1) - 0.06461 ) / 0.2458
X4 = ( INTERPRO - 1.075 ) / 1.410
X5 = ( MIM - 0.4325 ) / 0.6495
X6 = ( GBACC - 3.107 ) / 2.581
|
RMS=0.93, Mean_error=0.70 |
RMS=1.22, Mean_error=0.92 |
| SVM |
Default: SVM-Type: eps-regression;
SVM-Kernel: radial; cost: 1; gamma: 0.0625;
epsilon: 0.1; Number of Support Vectors: 75
|
RMS=0.58, Mean_error=0.42 |
RMS=1.02, Mean_error=0.80 |
| Linear Model |
Default: GCI = 4.13861 +0.03046*DBSNP +0.43011*ENSEMBL -0.03288*GBACC +0.11546*GO -0.06262*HOMOLOGENE
+0.26248*INTERPRO +0.23549*KEGG +1.02523*MEDLINE +0.27077*MIM +0.08164*NAME -0.07937*PDB
-0.11282*PRINTS -0.08908*PROSITE -0.04363*REFSEQ +0.56528*SPID +0.06029*TREMBL |
RMS=0.98, Mean_error=0.77 |
RMS=1.27, Mean_error=0.99 |
| Neural Nets |
size=3, skip=T, linout=T, maxit=100:
a 16-3-1 network with 71 weights, options were -
skip-layer connections, linear output units |
RMS=0.32, Mean_error=0.22 |
RMS=1.76, Mean_error=1.32 |
| Regression Tree |
Default: Used attributes:
HOMOLOGENE, MIM, MEDLINE, NAME |
RMS=0.95, Mean_error=0.72 |
RMS=1.23, Mean_error=0.95 |