Statistical Methods for Analyzing DNA and PamChip Microarray Data

In this research, we have addressed several questions on bioinformatics from statistical perspective. More specifically, we have developed several statistical methodologies to analyze spikes-in cDNA, peptide microarray data (based on PamGene technology) and human microRNA data (Alu-Repeats).For the preprocessing of spotted microarrays, different methods have been described. Overviews are given, e.g., by Leung and Cavalieri (2003), Quackenbush (2002), and Bilban (2002). In general, the preprocessing of spotted microarrays largely depends on the calculation of the log-ratios of the measured intensities. However, for some analyses, having access to absolute expression levels seems more suitable (Kerr et al., 2000). ANOVA models for absolute expression levels have been proposed, e.g., byWolfinger et al. (2001). We propose the use of external reference RNAs (also known as spike-in controls or spikes) to preprocess cDNA microarray data. We model the measured intensities as a function of the concentration in a statistical manner. This prompts us to use a preprocessing model proposed by Engelen et al. (2006) in the context of nonlinear mixed-effects models. In this way,we should be able to obtain the asymptotic prediction intervals for the estimated RNA sample for each gene. The novelty of this approach is that the uncertainty of the parameters of the preprocessing model can directly be ascertained. Furthermore, it also accounts for the various sources of variability associated with microarray experiments (Thilakarathne et al., 2009). Microarray experiments are subject to certain limitations, namely, slide-cost and time. The outcome is that the biological sample size and the technical replication are often very small that the application of conventional statistical techniques needs to be refined. To this end, we specifically target microarray experiments with a single intensity measurement per biological condition. Such datasets may arise in singlereplicated experiments (e.g. prokaryotes) or time course ones (Peeters et al., 2004) but they can also result from preprocessing methods (Thilakarathne et al., 2009). The model we proposed to handle this type of situation is a general one and one that can be developed to also handle more complex experimental designs (Thilakarathne et al., 2011a).MicroRNAs are 19 to 22 nucleotide long non-coding RNAs that influence gene expression by repressing translation or causing mRNA degradation (Bartel, 2004; Valencia-Sanchez et al., 2006). A recent study shows that the microRNA cluster on the human chromosome 19 (C19MC) is linked to Alu repeats which facilitated the expansion of C19MC. In this study, we try to show that the repeat elements in the microRNA cluster in human chromosome 19 are different from the repeat elements on the same chromosome. To test this, we compared the mean, median and standard deviation of the length of the repeat elements in C19MC to the distribution of the same characteristics in 1250 randomly selected windows of size 100Kb (Lehnert et al., 2009). Protein kinase plays an important role in oncology research. They are the enzymes that modify other proteins by adding a phosphate group (phosphorylation). Phosphorylation activates or deactivates a lot of protein enzymes, causing or preventing the mechanisms of diseases such as cancer and diabetes. A deregulated kinase activity can frequently cause diseases such as cancer. Therefore, inhibiting the kinase activity is important for controlling the activity of cancer cells. For instance, protein tyrosine kinases constitute a sizable class of drug targets in oncology (Versele et al., 2009). Identifying the responsive cell lines to a particular kinase inhibitor is therefore an important task in this genre of studies. In a protein kinase assay, the kinetic behavior of kinase can be monitored in real time using the PamGene technology (PamStation) for a set of cell lines. To analyze the PamChip microarray data, we propose a flexible semi-parametric mixed model. This approach would allow for the estimation of rate of phosphorylation (velocity) as a function of time, together with pointwise confidence intervals. Our model makes it possible to test whether the velocity of kinase inhibition differs from responding to non-responding cell lines. This can be tested at any time point and for entire time series profiles (Thilakarathne et al., 2011b).status: publishe