DNA microarrays are a high-throughput technology useful for functional genomics and gene expression analysis. While many microarray data are generated in sequence, most expression analysis tools are not utilizing the temporal information. Temporal expression profiling is important in many applications, including developmental studies, pathway analysis, and disease prognosis. In this paper, we develop a learning method designed for temporal gene expression profiling from massive DNA-microarray data. It attempts to learn probabilistic lattice maps of the gene expressions, which are then used for profiling the trajectories of temporal expressions of co-regulated genes. This self-organizing latent lattice (SOLL) model combines the topographic mapping capability of self-organizing maps and the generative property of probabilistic latent-variable models. We empirically evaluate the SOLL model on a set of cell-cycle regulation data, demonstrating its effectiveness in discovering the temporal patterns of correlated genes and its usefulness as a tool for generating and visualizing interesting hypotheses.
- Correlated genes
- DNA-microarray data
- Learning latent-variable models
- Temporal expression profiling
ASJC Scopus subject areas
- Artificial Intelligence