In this pre-print, the authors characterized the transcriptome of C. elegans at 30 timepoints during development. They used fPCA to cluster 14 co-expression modules from 19,711 genes and then fitted a cubic function to the expression profile of each module.
The authors then conducted hierarchical clustering on the parameter values of polynomial regression to relate co-expression modules. They found that related modules have a spatial bias on chromosomes. By comparing the evolutionary rates of genes, they also found that genes in one module (M4) are expressed in the early developmental stage but evolve rapidly. This discovery favors an evolutionary model according to which phenotypic constraint is relatively weaker during earlier and later time points in development.
We found the introduction exciting. The authors mentioned several evolutionary theories, which are mostly based on “old-school” morphological interpretations of phenotypes, and used molecular data to test some of them. We were also excited about how transparently the authors show their data. The visualization of the data was nice and well organized.
However, we found that the rationale behind the choice of the methods could have been more clearly justified. For example, is the cubic function the best approach to fitting the developmental expression patterns? We had some ideas about why the authors may have used this function, but it would have been helpful to know their justification. We were also curious as to whether the way they handle the data throws away a lot of information. For example, fPCA finds clusters that are linear combinations of two functions; it would have been helpful to see these functions. Because the genes that affect morphology are a small part of the genome, this method may lose rare but interesting biological patterns, especially when the number of modules it returns is low.
Although the module M4 is an interesting outliner from several aspects, we are wondering if it might be something unique about this data (e.g., a property of chromosome 2). This module is also enriched with genes of histones, but it is hard to understand how the evolution theories introduced in the introduction can be applied to the histones. We wonder what the results would look like if other methods were employed in the analysis, e.g., the recently published Dirichlet-process Gaussian-process infinite mixture model[McDowell et al, 2018].