Table 1.

comparison for unsupervised methods: Silhouette coefficients and number of genes for each cluster and unsupervised clustering method (no labels). Laplacian Eigenmaps+k-means leads to higher silhouette coefficients.

	k-means		PCA+k-means		LE+k-means
cluster	sil	# genes	sil	# genes	sil	# genes
1	0.0200	65	0.7329	126	0.6535	103
2	0.3067	146	0.6221	60	0.7049	125
3	0.4078	180	0.7002	168	0.6862	174
4	0.4068	234	0.6840	198	0.6848	154
5	0.3401	255	0.7423	157	0.7831	97
6	0.2960	252	0.7033	130	0.7949	389
7	0.3442	90	0.6795	126	0.7369	120
8	0.6509	9	0.6800	65	0.6953	270
9	0.3900	254	0.6393	190	0.7800	91
10	0.2162	34	0.7130	187	0.7046	79
11	0.3056	112	0.6517	182	0.7606	141
12	0.3531	165	0.7162	155	0.7487	122
13	0.4636	182	0.6925	117	0.9889	3
14	0.4267	167	0.7422	205	0.7118	125
15	0.6529	114	0.6968	184	0.5997	85
16	0.1593	86	0.5266	9	0.7214	236
17	0.5488	13	0.6792	84	0.6839	83
18	0.4323	253	0.6956	211	0.7380	135
19	0.1749	20	0.7151	118	0.6466	72
20	0.3076	133	0.6926	170	0.7243	121
21	0.4314	174	0.7041	115	0.7461	199
22	0.4394	130	0.7342	116	0.7442	275
23	0.4538	210	0.7252	192	0.6849	115
24	0.4366	138	0.6792	151	0.8534	102