Fig. 3

Mean run times and standard deviations for interaction analysis using R’s lm function, MatrixEQTL, and pulver. The execution times are in milliseconds. We fitted a line through the time points for each package. R’s lm function was very inefficient for this type of interaction analysis, and only the first two points are shown for every benchmark. Shown are four different panels (a-d). In panel a the number of columns of the matrix is set to 10, the matrix to 20 and the number of observations is set to 100, while the number of columns for the matrix is varied from 10 to 10,000. In panel b number of columns of the matrix is varied from 10 to 10,000 while the number of columns for the matrix is set to 10 column, the matrix to 20 column and number of observations is set to 100. In panel c the number of observations are varied from 10 to 10,000 while the number of columns for each matrix are fixed (all with 10 columns). In panel d number of columns of the matrix is varied from 10 to 10,000, while the number of columns of the matrix is set to 20, the matrix to 10 and the number of observations is set to 100