Refinement of the synthetic matrix wma. The sequence logos of the initial, DISPARE and width-optimized matrices are shown in A), B) and C). We compared the performances of these three reference models against four motif discovery tools: BayesMD, MEME, PREGO and WEEDER. The ROC performances of this assessment are summarized in D). The DISPARE discriminative refinement algorithm improved the initial matrix and obtained better results that the other methods.