Table 3.

Examples of Rules Induced for transcription

30MIN − 4H(Constant) AND 1H − 8H(Decreasing) AND 16H − 24H(Increasing) → Process(transcription)

0MIN − 1H(Constant) AND 30MIN − 4H(Increasing) AND 8H − 16H(Decreasing) AND 16H − 24H(Decreasing) → Process(transcription)

15MIN − 1H(Decreasing) AND 30MIN − 4H(Constant) AND 6H − 24H(Increasing) → Process(transcription)

A total of 5402 rules was generated for transcription. Using Michaelski's rule quality measure (Torgo 1993), we selected rules that cover the highest number of genes encoding proteins involved in transcription (cf. coverage) and that exhibit the highest ability to discern these genes from genes involved in all other processes (cf. accuracy). The rules originate from the known genes PBX3, ZNF222, and TRIP7, respectively. The first two rules also participated in the classification of unknown genes KIAA1799 and MGC5469, although only KIAA1799 received a fraction of votes high enough to be classified as transcription.

The three rules shown in the Table had a Michalski's value of 0.75 (μ × accuracy + (1 − μ) × coverage; where μ = 0.5 + 0.25 × accuracy). There was a total of 50 rules for transcription with Michalski's value 0.75. Intervals where gene expression profiles fit the template for “increasing” are shown in bold letters, whereas intervals that fit the template for “decreasing” are shown in italics (for a description of “increasing/decreasing” templates see Methods).