Sequential prediction bounds for identifying differentially expressed genes in replicated microarray experiments

Sequential stopping for high-throughput experiments

In high-throughput experiments, the sample size is typically chosen informally. Most formal sample-size calculations depend critically on prior knowledge. We propose a sequential strategy that, by updating knowledge when new data are available, depends less critically on prior assumptions. Experiments are stopped or continued based on the potential benefits in obtaining additional data. The underlying decision-theoretic framework guarantees the design to proceed in a coherent fashion. We propose intuitively appealing, easy-to-implement utility functions. As in most sequential design problems, an exact solution is prohibitive. We propose a simulation-based approximation that uses decision boundaries. We apply the method to RNA-seq, microarray, and reverse-phase protein array studies and show its potential advantages. The approach has been added to the Bioconductor package gaga.

Histone deactylase 1 expression is increased in the prefrontal cortex of schizophrenia subjects: analysis of the National Brain Databank microarray collection

Histone deactylase enzymes are responsible for the deacetylation of histone tails, and consequently influence gene regulation through their ability to modify chromatin structure surrounding promoter regions. We analyzed the microarray collection of the National Brain Databank to investigate differential expression of these enzymes in the prefrontal cortices of control, schizophrenia and bipolar subjects. HDAC1 expression levels were significantly higher in schizophrenia versus normal subjects. The mRNA expression level of an epigenetically regulated schizophrenia candidate gene GAD67 was strongly and negatively correlated with the mRNA expression levels of HDAC1, HDAC3 and HDAC4 levels. These findings provide additional support for the proposal that epigenetic factors are operative in the brain pathology of patients with schizophrenia.

Selecting differentially expressed genes using minimum probability of classification error

Discovery of differentially expressed genes between normal and diseased patients is a central research problem in bioinformatics. It is specially important to find few genetic markers which can be explored for diagnostic purposes. The performance of a set of markers is often measured by the associated classification accuracy. This motivates our ranking of genes depending on the minimum probability of classification errors (MPE) for each gene. In this work, we use Bayesian decision-making algorithm to compute MPE. A quantile-based probability density estimation technique is used for generating probability density functions of genes. The method is tested on three datasets: colon cancer, leukaemia, and hereditary breast cancer. The quality of the selected markers is evaluated by the classification accuracy obtained using support-vector-machine and a modified naive Bayes classifier. We obtain 96.77% accuracy in colon cancer and 97.06% accuracy in leukaemia, using only five genes in each case. Finally, using just three genes we get 100% accuracy in hereditary breast cancer. We also compare our results with those using the genes ranked by p-value and show that the genes ranked by MPE perform better or equal to those ranked by p-value.