Figure 1.

Flow chart of the quadratic regression method. The gene selection and pattern classification procedure of our quadratic regression method. y_ij is the expression level; x is time; β_0j, β_1j, and β_2j are the parameters of intercept, linear effect, and quadratic effect, respectively; ε_ij is the random error. Among the 9 regression patterns, FLAT stands for no statistically significant differential expression over time; LU stands for linear up; LD stands for linear down; QC stands for quadratic concave; QV stands for quadratic convex; QLCU stands for quadratic linear concave up; QLCD stands for quadratic linear concave down; QLVU stands for quadratic linear convex up; QLVD stands for quadratic linear convex down.