A Model for Temporal Abstraction in Gene Expression Studies

When utilizing information from increasingly voluminous biomedical and genomic databases into actionable data for health care, treatment of temporal data still remains a challenge. Frequently temporal research is based on stimulus response studies and includes searching for temporal effects or time patterns in gene sets. Digital gene expression (DGE) technologies like rna-seq seem to replace microarray technologies in the near future for many functional genomics applications. This study explores the feasibility of searching for temporal patterns based on knowledge-based temporal abstractions. Those imply conversion of expression values into an interval-based qualitative representation expressing amount of change over time. The amount of change is determined by statistical significance. For microarray studies one approach uses Bioconductor limma software modelling the normalized intensities in the framework of the linear model. Empirical Bayes methods result in a moderated t-statistic that reduces the pooled variance by borrowing information across all genes. We use the moderated paired t-test to determine significant differences in consecutive time points. While this approach assumes that the experiment is based on one particular platform, comparison across platforms can be done by comparing p-values. Therefore, in our model the p-values and the direction of the change inform the temporal abstraction. We discuss this approach in the framework of our SPOT software.