Protein surface representation and analysis by dimension reduction

Table 1 Performance of dimension reduction methods for mapping a triangulated hemisphere

Methods	Area	Neighbor	Runtime (sec)
PCA	0.28	0.46	0.002
LLE	0.28	0.47	0.03
Laplacian	0.11	0.29	0.02
LLC	-0.01	0.02	1.16
AutoEncoderEA	0.01	0.11	2.83
SNE	0.72	0.53	6.81
SymSNE	0.03	0.01	6.75
CFA	0.27	0.12	4.36
GPLVM	0.28	0.46	0.3
NPE	0.28	0.47	0.03
LPP	-0.007	0.17	0.004
LLTSA	0.28	0.46	0.029
NCA	0.28	0.09	3.1
MCML	0.28	0.46	0.79
LDA	0.28	0.37	0.005
FactorAnalysis	0.24	0.21	0.004
tSNE	0.16	0.53	0.61
Isomap	0.81	0.53	0.29
LandmarkIsomap	0.69	0.53	0.06
ProbPCA	0.28	0.48	0.11
KernelPCA	0.04	0.23	0.007
MDS	0.28	0.46	0.004
DiffusionMaps	0.3	0.48	0.006
Sammon	0.73	0.54	0.1
Sinusoidalcartography	0.10	0.45	0.003

For each performance criteria, the top comparable scores are shown in bold. knn = 3 was used for the neighbor scoring.

ISSN: 1477-5956