Case Study III: Hybrid PCA-NN Algorithms for Continuous Microarray Time Series

In this chapter, we describe about the hybrid models of the principal component analysis (PCA) and neural network (NN) for the continuous microarray gene expression time series. The main contribution of our work is to develop a methodology for modeling numerical gene expression time series. The PCA-NN prediction models are compared with other popular continuous prediction methods. The proposed model can give us the extracted features from the gene expressions time series with higher prediction accuracies. Therefore, the model can help practitioners to gain a better understanding of a cell cycle, and to find the dependency of genes, which is useful for drug discoveries. Based on the results of two public microarray datasets, the PCA-NN method outperforms the other continuous prediction methods. In the time series model, we adapt Akaike's Information Criteria (AIC) tests and crossvalidation to select a suitable NN model to avoid the over-parameterized problem.

The outline of this chapter is as followed. In Section 6.1, we describe the background, like the neural network and the transformation algorithms, and their respective applications in the microarray analysis. In Section 6.2, we talk about the motivation for developing the PCA-NN algorithm. In Section 6.3, it is the data description of the public datasets used in our study. In Section 6.4, we describe the details of our proposed methodology and the result comparison of the different methods. In Section 6.5, we discuss about the results of our system and further integration that can be developed, basing on our experimental results.