Distinct and predictive histone lysine acetylation patterns at promoters, enhancers, and gene bodies.
Bottom Line: Unexpectedly, we found that histone acetylation alone performs well in distinguishing these unique genomic regions.Further, we found the association of characteristic acetylation patterns with genic regions and association of chromatin state with splicing.Taken together, our work underscores the diverse functional roles of histone acetylation in gene regulation and provides several testable hypotheses to dissect these roles.
Affiliation: Ludwig Institute for Cancer Research, 9500 Gilman Drive, La Jolla, California 92093-0653 Bioinformatics and Systems Biology Program, University of California, San Diego, La Jolla, California 92037 Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139.Show MeSH
Mentions: We asked to what extent we could predict retention of exon–intron junctions based on chromatin modifications as input features. We defined the constitutive class of exon–intron boundaries as those that have the maximum possible value of inclusion ratio, 0.999, in all transcripts of which they are a part. We defined two categories of alternatively spliced exon–intron boundaries based on their contribution to splice-site usage: group I class of boundaries comprising IR, AD (5′ end), and AA (3′ end) contribute negatively to splice-site usage, whereas group II class of boundaries comprising CA contributes positively to splice-site usage, as defined above. Using all 24 modifications, we obtained a maximal classification accuracy of ∼70% and AUC of 0.75 for group I exon–intron boundaries in IMR90 (Figure S6B, black). Although this is clearly greater than expected at random, we asked if we could further improve the classification accuracy by taking into consideration other factors. For instance, exon–intron boundaries within close proximity of each other may share the same chromatin signature, which would cause difficulty in classification. To verify this, we filtered any retained exon–intron boundary within different distances of the constitutive exon–intron boundaries and found a steady improvement in accuracy of classification with filtering distance (Figure S6B, black to red). Now, if we consider filtering the group I elements for any constitutive exon–intron boundaries, we actually observed a worsening of the performance (Figure S6B, black vs. dotted blue). We obtained the best possible accuracy of classification with an AUC of 0.84 and maximal accuracy of 77.1% by using a filtering distance of 10 kb for determining the set of distal constitutive exon–intron boundaries in IMR90 (Figure 6A, blue). In 1H, we observed the same trend (data not shown) and obtained a maximal accuracy of 76.5% and AUC of 0.84 for classification of group I exon–intron junctions against distal constitutive ones (Figure 6B, blue).
Affiliation: Ludwig Institute for Cancer Research, 9500 Gilman Drive, La Jolla, California 92093-0653 Bioinformatics and Systems Biology Program, University of California, San Diego, La Jolla, California 92037 Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139.