PROMOTER STRENGTH PREDICTION BASED ON OCCURRENCE FREQUENCIES OF CONSENSUS PATTERNS

A training sample of 14 sequences of E. coli, each 70 or 69 base pairs long, with their (relative) promoter strengths given by Deuschle et a l. (1986, EMBO J. 5, 2987-2994), is used to find a relation between th e occurrence frequencies of the two consensus patterns and the promote r strength. The approach is restricted to an analysis of purine and py rimidine organization using the theory of stationary alternate Markov chains of first order. Further, it is shown, both empirically by regre ssion analysis, and by a Markov-chain-oriented statistical analysis, t hat the difference of occurrence frequencies and the determinant of tr ansition matrix, which was introduced in a previous paper (Recknagel e t al., 1993, J. theor. Biol. 162, 75-80), are equivalent measures with respect to the task of promoter strength prediction. An empirical reg ression equation is given that allows the promoter strength to be fore cast from the occurrence frequencies of the canonical hexamers in the consensus boxes. Three E. coli promoters of an examination sample, sep arated from the training sample, are classified this way, in agreement with the experimental findings.