Joint differential expression estimation for spatial and temporal data
This work was motivated by a spatial (16 brain regions) and temporal (15 time periods) human brain transcriptome data set, where the challenge is to model the complex data structure. We developed a two-step inferential procedure to detect expressed genes, and differentially expressed (DE) genes between adjacent time periods. Markov Random Field (MRF) priors have been implemented to capitalize on the temporal and spatial similarity. Compared with simple marginal analysis, our joint model identifies an order of magnitude more genes under appropriate FDR control, and the DE genes tend to have small but consistent changes among the brain regions. Genes carrying high risk for autism, tend to be differentially expressed, especially during periods when cognitive and social skills are developed. This work has been published in the Annals of Applied Statistics. For the MRF model, we implemented a Monte-Carlo Expectation-Maximization (MCEM) algorithm to estimate the parameters. In a follow-up paper, I extended the model to discrete data and applied it to a mouse brain development RNA-Seq dataset, where a much more efficient EM algorithm (~100 fold faster with similar accuracy) with mean field-like approximation was implemented. This work has been published in Statistical Applications in Genetics and Molecular Biology.
- Lin Z, Li M, Sestan N, and Zhao H: A Markov random field-based approach for joint estimation of differentially expressed genes in mouse transcriptome data. Statistical applications in genetics and molecular biology. 2015, 15 (2): 139-150. [paper link] [software link]
- Lin Z, Sanders SJ, Li M, Sestan N, State MW and Zhao H: A Markov Random Field-based approach to characterizing human brain developments using spatial-temporal transcriptome data. Annals of Applied Statistics 2015, 9 (1): 429-451. [paper link]