Gene expression based prototype for automatic tumor prediction

Background

Automatic detection of tumors is a challenging task due to the heterogeneous phenotypic and genotypic behaviors of cells within tumor types [1–3]. In recent years, a number of research endeavors have been reported in literatures that exploit microarray gene expression data to predict tissue/tumor types with high confidence [3–14]. However, in predicting tissue types, the above mentioned works neither explicitly considered correlation among the genes nor the probable subgroups within the known groups. In this work, our primary objective is to develop an automated prediction scheme for tumors based on DNA microarray gene expressions of tissue samples.

Material and methods

The workflow to build the tumor prototypes is shown in Fig. 1. Considering various sources of variation in array measures, we estimate tumor-specific gene expression measures using a two-way ANOVA model. Then, marker genes are identified using Wilcoxon [15] and Kruskal-Wallis [16] test. We then group the highly correlated marker genes together. Then, we obtain eigen-gene expressions measures [10] from each individual gene group. At the end of this step, we replace the gene expression measurements with eigen-gene expression values that conserve correlations among the strongly correlated genes. We then divide the tissue samples of known tumor types into subgroups. The CS measure [17] is exploited to obtain the optimal number of gene groups and tissue subgroups within each tissue type. The centroids of these subgroups of tissue samples represent the prototype of the corresponding tumor type. Finally, any new tissue sample is predicted as the tumor type of the closest centroid.

Figure 1

Simplified workflow to build the tumor prototypes.

Results

To evaluate the proposed tumor prediction scheme, five different gene microarray datasets [3–5, 7–9] are used, all of which were obtained using Affymetrix technology. We use leave-one-out cross validation method. Table 1 shows a summary of our experimental results for all the datasets. We provide relevant intermediate results along with the final classification accuracy. Finally, Table 2 shows the performance comparison between our proposed prediction scheme and the methods discussed in original works [3, 5, 7–9] wherein the corresponding datasets are published. We also compare our classification accuracies with those of a Supervised Clustering method [4] for completeness.

Table 1 Experimental results with different dataset.

Table 2 Comparison of methods.

Conclusions

In this work, we propose a novel, seamless, and integrated technique of automatic tumor detection using Affymetrix microarray gene expression data. We appropriately normalize the data by estimating tumor-specific gene expression measures using an ANOVA model. Furthermore, our novel tumor prediction scheme explores molecular information such as probable correlations among genes and probable unknown subgroups within known tumor types. We demonstrate the efficacy of our proposed scheme using five different Affymetrix gene expression datasets.