PCA-maSigFun [4] identifies the major(s) gene expression changes within each functional class and evaluates whether these changes are significantly associated to the time. In summary, PCA is applied to the gene-expression submatrix associated to the genes belonging to each functional category. The scores of the relevant principal component(s) of these PCAs are taken as joined expression profile(s) for the functional class. The regression based time-course analysis methodology maSigPro [1] is then applied to the joined profiles (PC scores) to identify function-related subset of genes with expression changes significantly associated to the time. Note that each functional class can result in more than one joined profile when more than one subset of correlated genes exist within that functional category. that we consider that a functional block might contain several patterns of coordinative gene expression. This program returns lists of significant functional classes (an their representative joined profiles) for each of the series included in the experiment.

Parameters for PCA-maSigFun gene selection:

• Data: txt file with expression data, genes in rows, arrays in columns. The file must contain an additional row with arrays names and a column with gene names.
  

• Covariates: txt file with experimental design information, containing as many columns as arrays and as many rows as experimental factor. Each cell contains the value of the array in the experimental factor. E.g:

• Quantitative factor: name of the numerical variable of the experimental design, normally the time.
• Qualitative factor: name of the categorical variable of the experimental design.
• Control group: name of the reference series in the regression model (in the example is Ctr).
• Annotations: the annotations can be uploaded in a two columns (gene tab annotation) txt file.
• Polynomial degree: degree for the regression model. The maximum allowed degree is #time_points – 1.
  
• Alpha: significant level for gene selection.
  
• R-Squared cut-off: required level of the goodness of fit of the regression model. This parameter is between 0 and 1. Higher values indicate well fitted models. We recommend values between [0.4,0.8].
  
• Cut-off: Variability level to select Principal Components in each category.
• Selection factor: criterion to select components can be:
  • Proportion of acumulated variability. Posible cut-off values are in (0,1).
  • Proportion of variability of each PC. Posible cut-off values are in (0,1).
  • Average: components are selected that explain more than “cut-off” times the average component variability. The recommended “cut-off” values are in [1,1.5].

Parameters for PCA-maSigFun visualization
To show the trajectories in plots maSig visualization applies clustering methods to group functional categories with similar trends and summarize the graphical display.

• Series to see: name of the series to visualize from the available series.
  
• Clustering method: available methods are:
  • 'hclust': hierarchical clustering
    
  • 'kmeans': k-means
    
• Number of clusters: groups to split gene selection to show results.

PCA parameters: Threshold for significant gene contribution for the PCA model. This threshold allows the identification of the genes that most contribute to the selected components. It can be computed by applying several procedures:

• Resampling: where a null Leverage distribution is created by permuting columns of expression data and genes are selected at the “alpha” percentile of the null distribution.
• minAS: where a density function is calculated on the data and genes are selected on a local minimum basis [7].
• Gamma: where a gamma distribution is adjusted to the distributions of the gene loadings, and genes are selected at the “alpha” percentile of the gamma distribution [7].
• Custom: where the user can decide the threshold.