Figure 1.

The fREDUCE algorithm. A set of possible IUPAC strings are generated from the input sequence. For each IUPAC string, we compute a pseudo-Pearson coefficient, which is an estimate and upper bound on the true Pearson coefficient. After the vast majority of motifs are filtered out using the pseudo-Pearson value, we then compute true Pearson coefficients for the remaining motifs and select the top motif. The residual expression value is then used to iteratively derive subsequent motifs.