SOMBRERO manuscript; supplementary information

Supplementary Table 1.Accuracy of SOMBRERO (using default settings outlined in the manuscript) compared to SOMBRERO using random PWM initialization and a K-means algorithm with the same number of cluster centres as SOM nodes in each case. The results obtained using random initial PWMs do not significantly differ from those obtained using the default initialization. In contrast, the K-means clustering algorithm performs significantly worse. There results show that SOMBRERO is not sensitive to perturbations in the initial PWM settings, and also show that the neighbourhood update equation (which basically distinguishes the SOM from the K-means algorithm) is an important factor in SOMBRERO’s improved performance.

Supplementary Table 2. The motif finding power of the initial grid of PWMs before the training phase. The grid of PWMs, initialized as outlined in the section "Initializing the SOM", was used to find motifs before any SOM training steps were carried out. The test was carried out using the Yeast datasets, and the best match to the real motif in the top 10 results is reported. Each motif found in this test is basically a random grouping of l-mers according to their GC% content. Therefore it is not surprising that some true positive matches should be found in this fashion, and it is also not surprising that the FP rate is so high in each case.

• Disclaimer
• Privacy
• Copyright
• Accessibility