Doctoral Thesis Defense: Characterizing Pathway-Specific Anomalies in Temporal Patterns of Gene Expression

November 19, 2019
3:00 PM
Halligan 102
Speaker: Michael Pietras
Host: Donna Slonim


Understanding the dynamic aspects of molecular processes is essential, especially for inherently temporal functions such as those involved in development, aging, or progressive disorders. In previous work, our group developed the FRaC anomaly detection algorithm and used it to identify anomalous mRNA expression patterns and to characterize individual anomalies by identifying dysregulated molecular functions. However, FRaC operates by training supervised models for each feature in a data set, and scaling to substantially larger data sets would require prohibitive amounts of computation time and memory. We show that by using information provided by preexisting knowledge of biologically relevant pathways, it is possible to substantially reduce usage of computational resources while preserving the same anomaly detection accuracy.

However, static analysis of many expression data sets, including that provided by FRaC, is not enough to fully capture time-related patterns of dysregulation that might be present highly time-related diseases or disorders such as autism, Huntington's disease, or Alzheimer's disease. To characterize temporal dysregulation in these and other disorders, we develop TEMPO, a pathway-based outlier detection approach for finding pathways showing significant temporal changes in gene expression patterns in data sets with additional temporal annotation. Our experiments demonstrate that a temporal pathway approach can identify new functional, temporal, or developmental processes associated with specific phenotypes.

Finally, we develop aTEMPO, combining the two approaches developed with TEMPO and Scalable FRaC by using an anomaly detection model and virtual time series to identify anomalous temporal processes in specific disease states. We demonstrate that this approach provides superior anomaly detection results when compared to anomaly detection approaches that do not explicitly consider a temporal component, and that aTEMPO can informatively characterize individual patients and suggest personalized therapeutic approaches.