Fig. 5

Unsupervised clustering of CGIs identifies distinct chromatin landscapes. A CGIs were clustered based on their percentage methylation and ATAC-seq signal (RPKM) in the Holstein Friesian data. Clustering was performed using finite Gaussian mixture modelling (GMM) fitted by the expectation-maximisation (EM) algorithm. Each CGI was only assigned to one gene which had the closest TSS. The RNA-seq expression values for the nearest gene to each CGI are shown as well as the distance of each CGI from the nearest TSS. Clusters are ordered by increasing median percentage methylation and are numbered according to this order. This clustering was repeated for the N’Dama and Nelore data (see Additional file 1: Fig. S5). B Circos plot showing the degree of CGI overlap across different clusters between breed pairs. Precedence is given as follows: Holstein Friesian > N’Dama > Nelore. The outermost labels indicate the breed data used in the clustering analysis followed by the cluster number. The outermost bars show the relative overlap of CGIs in each breed that fall within the clusters of the other two breeds. This is also shown in the innermost and middle bars for the Holstein Friesian clusters and Nelore clusters respectively. Specifically for the N’Dama clusters, the innermost bar shows the contribution of CGIs to the Nelore clusters, while the middle bar shows the contribution of CGIs to the Holstein Friesian clusters. The ribbon colours represent different cluster numbers and the ribbon size is equivalent to the proportion of CGIs within a cluster. Bcell = B cell, CD4 = CD4 T cell, CD8 = CD8 T cell, gdT = γδ T cell, NK = NK cell, mono = monocyte and gran = granulocyte