Skip to main content
Figure 1 | Journal of Experimental & Clinical Cancer Research

Figure 1

From: Whole blood transcriptome correlates with treatment response in nasopharyngeal carcinoma

Figure 1

Data Analysis Outline. (a) Microarray gene profiling raw data were pre-processed for quality control before analysis. First, all samples were normalized using MAS5 algorithm and only probes flagged as “present” were retained. The “present” probes were then compared with the list generated in MAQC studies for Affymetrix Human U133 plus 2; non-overlapped probes were deemed unreliable and, therefore, excluded. Expression values of each probe were then logarithm transformed. (b) Logistic regression multivariate analysis of the gene expression values was performed to evaluate the AUC of each gene and of different multi-gene combinations. Significance of associations between gene expressions was determined using a logrank test. The best set of coefficient values that maximize the separation between the positive and negative groups were determined. Later, the log ratio calculation was determined in order to reduce the impact of possible noise (c). Thresholds were then set to evaluate sensitivity, specificity and the stability of the prediction. Two individual genes were combined to form a gene pair (d). Then the single pair of genes was coupled to form 2-pair and then 3-pair gene combinations. Logistic regression values were calculated for each gene pair, and we showed that in each case when genes were combined, the area under the curve (AUC ROC) increase.d

Back to article page