Temporal Gene Expression Patterns

In a previous study [11], the authors established some relationships between temporal gene expression patterns of 112 rat CNS (Central Nervous System) genes and the development process of the rat's CNS. Three major gene families were considered: Neuro-Glial Markers family (NGMs), Neurotransmitter Receptors family (NTRs) and Peptide Signaling family (PepS). All other genes measured in this study were lumped by the authors into a fourth family: Diverse (Div). All families were further subdivided by the authors, based on apriori biological knowledge. Gene expression patterns for the 112 genes of interest were measured (using RT/PCR: [16]) in cervical spinal cord tissue, at nine different developmental time points. This yielded a $112 \times 9$ matrix of gene expression data. To capture the temporal nature of this data, the authors transformed each (normalized) 9-dimensional expression vector into a 17-dimensional vector, including also the 8 difference values between expression levels in successive time points. This transformation emphasizes the similarity between genes with closely parallel, but offset, expression patterns. Euclidean distances between the augmented vectors were computed, yielding a $112 \times 112$ distance matrix. Next, a phylogenetic tree was constructed for this distance matrix (using the FITCH program [4]). Finally, cluster boundaries were determined by visual inspection of the resulting tree. Some correlation between the resulting clusters and the apriori family information was observed. The CAST algorithm was tried on the same data in the following way: The raw expression data was preprocessed in a similar manner: first the normalized expression levels were augmented with the derivative values. Then, a similarity matrix was computed based on the L₁ distance between the augmented 17-dimensional vectors. The CAST algorithm was applied to the similarity matrix. Clusters were directly inferred (figure 12.8).

  

Figure 12.8: The unprocessed data is compared to the output of the clustering algorithm. Top: The similarity matrix of the unprocessed data, compared against the new permutation according to the found clusters. Bottom: The raw gene expression matrix is ordered according to the permutation produced by the clustering algorithm and compared to the original order.