Potato Stress Example


We start the analysis by exploring data from a global perspective using the ASCA-genes module. After loading the example data, we select default option values. By Include interaction between factors, we allow the model to analyse TimexTreatment interactions and by Join interaction with second factor, we study the Treatment and the TimexTreatment effects together. Once the job is completed we can visit the results page by selecting this on the Job panel. The Variability analysis of this dataset reveals (54.13%) of the variation in the dataset is associated to the TimexTreatment factor while a lower percentage (9,26%) is due to solely the Time factor.
These variations can be summarized by PCA submodels with one and two components, respectively. Two components of the TimexTreatment submodel explain 72% of the variation associated to this factor.
The first component describes the trajectory showed in the right graph of Figure 1. It describes different responses between Salt/Cold and Control/Heat treatments, starting differences at time point 3, peaking at 9 hours and being maintained until the end of the experiment.
By looking at the leverage/SPE plot for Time-Treatment submodel we observe that most genes have values close to 0 at both parameters and only some of them present high values and are in the selection areas (Figure 1). Genes in the red region are well-modelled genes, as they have high leverages, while genes on the blue area are odd- or bad-modelled genes, as these present high SPE values.

Figure 1

Figure 1: Leverage-SPE plot and trajectory plot for Time-Treatment submodel.


The first output of the ASCA-Functional module is a summary of the ASCA model (also showed in the ASCA-genes module) to explore the general trends.
ASCA-Functional applies Fatiscan (GSA analysis) to the loadings of the genes of each component selected in the ASCA model. In our example we have 1 component for model Time and 2 components for submodel TimexTreatment, so 3 Fatiscan analyses will be developed for the ASCA-functional module. As previous ASCA-genes results suggest that the major transcriptional pattern is the first component of submodel TimexTreatment, depicted in Figure 1, we focus on this GSA analysis.
The output offers a table with the functional categories, the annotated genes in this category on the top of the partition (List 1: upregulated genes), the annotated genes on the bottom of the partition (List2: down-regulated genes), p-value of the Fatiscan test and the adjusted p-value.
Original result
The categories can be sorted by the adjusted p-values to select the most significant categories.
Sorted result
In this example a significant number of enriched functional categories were found. Processes associated to upper rank positions (up-regulated in Cold/Salt stresses) relate to protein synthesis and degradation, lignin biosynthesis, diverse hormone signaling pathways and response to several stimuli. Photosynthesis, microtubule-based movement, RNA binding and lypoxigenase activity were functions associated to bottom rank genes, i.e. were down-regulated at these stress conditions.


Having obtained a global impression on the transcriptional effects of the three abiotic stressors on potato seedlings we can next address a gene-wise analysis of serial changes with the maSigPro approach. After loading data in the maSigPro module, we choose as quantitative factor the Time, as qualitative factor the Treatment and as reference series Control. Default parameters indicate that second degree regression model will be applied and that genes will be selected at significance level of 0.05 and a R-Squared value above 0.7.
1315 genes were found to be statistically significant. SEA offers this result classifying significant genes by series according to the profile contrast considered. These lists are provided as text files ready for download:

We can now red-direct the maSigPro R object to the maSigVisualization module to create plots of the significant gene profiles. On the maSigVisualization tab, the maSigPro result is parsed we are asked to select the list of significant genes (Series to see) to visualize. We can for example, choose the ColdvsControl series and cluster significant genes in 9 groups by hierarchical clustering (hclust). Figure 2 shows the typical output of maSigPro visualization. For each cluster a trajectory plot is generated that summarizes gene expression along time and at each treatment for the genes belonging to that group. We can observe in this figure that most and more populated clusters indicate differences between Cold/Salt and Heat/Control treatments, the main transcriptional behavior already detected by ASCA-genes. In addition, other type of genes are revealed by this method. For example, cluster 7 collects genes with an exclusive early up-regulation upon Cold treatment.
Figure 2

Figure 2: Trajectory plots of 9 clusters created from genes with differences between Cold and Control (1300 genes).


By running maSigFun (degree 2, R-Squared 0.4, alpha 0.05) we identified 10 functional categories with a consistent expression patterns and significant differences between treatments. As in maSigPro this is offered by SEA classifying the categories by series:

We can also send to the maSig Visualization module these results to obtain graphic visualization of the trends of these functional categories.


Finally we run PCA-maSigFun to go a step deeper in the functional analysis. We selected degree=2, R-squared=0.7, Method for selection factor=%of that PC and cutoff=0.7. This tool indicated that a total of 37 functional groups had subsets of genes with highly correlated significant profile changes. Out of these categories have statistically significant differences for the experimental groups:

  • 3 categories with changes for the Control series,
  • 37 (all) categories with differences between Cold and Control,
  • 3 categories with differences between Heat and Control and
  • 31 categories with differences between Sald and Control groups.

maSigVisualization module for PCA-maSigFun offers two outputs (choosing 4 clusters and custom threshold=0.2):

  1. Output 1: Trajectory plots of clusters of categories, as in maSigFun.
  2. Output 2:A pdf file with the trajectory plots of each category next to a barplot with the correlation of each gene to the displayed trend.

The second output helps to understand the expression profile of the class and the participation of the class members in this profile. Figure 3 shows an example, the graphical output for glutamate metabolic process. On the left panel, we observe the trajectory plot for the class that again reveals the significant treatment differences already discussed. On the right panel, the gene loadings bar-plot indicates that 5 out of the 8 members of this class follow significantly the expression pattern displayed on the left, four of them with positive correlation and one with negative correlation.

[Figure 3]

Figure 3: Example of the output of PCA-maSigFun for glutamate metabolic process.
examples.potato.txt · Last modified: 2010/05/05 16:55 by mjnueda
CC Attribution-Noncommercial-Share Alike 3.0 Unported
www.chimeric.de Valid CSS Driven by DokuWiki do yourself a favour and use a real browser - get firefox!! Recent changes RSS feed Valid XHTML 1.0