Regulatory potential scores from genome-wide three-way alignments of human, mouse, and rat

Abstract

We generalize the computation of the Regulatory Potential (RP) score from two-way alignments of human and mouse to three-way alignments of human, mouse, and rat. This requires overcoming technical challenges that arise because the complexity of the models underlying the score increases exponentially with the number of species. Despite the close evolutionary proximity of rat to mouse, we find that adding the rat sequence increases our ability to predict genomic sites that regulate gene transcription. A variant of the RP scoring scheme that accounts for local variation in neutral mutational patterns further improves our predictions.

Figure 1

Cumulative distributions for two-way (blue) and three-way (red) RP scores on C(W)_REG and C(W)_AR. The text box contains the corresponding misclassification rates from leave-one-out cross-validation. Scores are all length normalized to correct for the slightly different sizes of the training segments (all around W = 100 bp).

Figure 2

Plots of two-way RP, three-way RP, and three-way LRP scores superimposed to annotations from the UCSC human genome browser along two selected genomic regions. (A) Approximately 10 kb around the cardiac α actin locus on human chromosome 15. (B) Approximately 10 kb around the C/EBPα locus on human chromosome 19. Scores are all length normalized to the fixed size W = 100 bp of the sliding windows on which they are calculated.