Evaluation of normalization methods for cDNA microarray data by k-NN classification

BMC Bioinformatics

Table 6 IMPROVEMENT of location normalization methods. IMPROVEMENT is defined (in the Results) based on improvement of LOOCV classification error rate of a given normalization method over that of N ONRM. The methods are arranged in the same order as those in Table 5. For a given data set, the biggest IMPROVEMENT(s) is shown in bold. The methods and data sets are described in Tables 1, 2 and 4, respectively.

Location Normalization method	IMPROVEMENT (%, LIVER CANCER)	IMPROVEMENT (%, LYMPHOMA)	IMPROVEMENT (%, RENAL CELL CARCINOMA)	IMPROVEMENT (%, GASTRIC CARCINOMA)	IMPROVEMENT (%, LUNG CANCER)	IMPROVEMENT RANGE (%)
N ONRM	0	0	0	0	0	0 - 0
GMEDIAN	19	7	33	22	5	5 – 33
SL LOESS	33	-2	44	56	3	-2 – 56
SL FILTERW3	23	19	44	45	0	0 – 45
SL FILTERW7	29	5	44	34	9	5 – 44
IG LOESS	34	30	44	33	5	5 – 44
IL LOESS	46	42	44	33	23	23 – 46
IST SPLINE	36	33	52	56	7	7 – 56
IG SG LOESS	33	51	44	100	21	21 – 100
IG LOESS-SL LOESS	44	56	50	56	33	33 – 56
IL LOESS-SL LOESS	48	58	44	44	26	26 – 58
IG LOESS-SL FILTERW3	22	49	61	33	33	22 – 61
IG LOESS-SL FILTERW7	44	58	50	56	40	40 – 58
IST SPLINE-SL LOESS	40	62	50	33	47	33 – 62
IST SPLINE-SL FILTERW3	22	48	61	34	42	22 – 61
IIST SPLINE-SL FILTERW7	42	52	44	34	42	34 – 52
Q SPLINEG	22	28	59	100	23	22 – 100
QS PLINER	18	54	59	100	23	18 – 100
Q SPLINEG-SL LOESS	32	26	50	78	37	26 – 78
Q SPLINEG-SL FILTERW3	29	30	27	78	12	12 – 78
Q SPLINEG-SL FILTERW7	26	28	55	78	37	26 – 78
QS PLINER-SL LOESS	23	61	56	78	37	23 – 78
QS PLINER-SL FILTERW3	23	58	39	78	9	9 – 78
QS PLINER-SL FILTERW7	16	56	50	67	23	16 – 67

ISSN: 1471-2105